some more menu options...

The Digest type needs more thought and is not necessary at the moment.

.dockerignore

@@ -1,8 +1,9 @@
 .vscode
 ollama
 app
+macapp
 dist
 llm/llama.cpp
 .env
 .cache
-test_data
+test_data

.gitattributes

@@ -0,0 +1 @@
+llm/ext_server/* linguist-vendored

.github/ISSUE_TEMPLATE/10_bug_report.yml

@@ -0,0 +1,60 @@
+name: Bug report
+labels: [bug]
+description: Something isn't working right.
+body:
+  - type: textarea
+    id: description
+    attributes:
+      label: What is the issue?
+      description: What happened? What did you expect to happen?
+    validations:
+      required: true
+  - type: dropdown
+    id: os
+    attributes:
+      label: OS
+      description: Which operating system are you using?
+      multiple: true
+      options:
+        - Linux
+        - macOS
+        - Windows
+        - Docker
+        - WSL2
+    validations:
+      required: false
+  - type: dropdown
+    id: gpu
+    attributes:
+      label: GPU
+      description: Which GPU are you using?
+      multiple: true
+      options:
+        - Nvidia
+        - AMD
+        - Intel
+        - Apple
+        - Other
+    validations:
+      required: false
+  - type: dropdown
+    id: cpu
+    attributes:
+      label: CPU
+      description: Which CPU are you using?
+      multiple: true
+      options:
+        - Intel
+        - AMD
+        - Apple
+        - Other
+    validations:
+      required: false
+  - type: input
+    id: version
+    attributes:
+      label: Ollama version
+      description: What version of Ollama are you using? (`ollama --version`)
+      placeholder: e.g., 0.1.32
+    validations:
+      required: false

.github/ISSUE_TEMPLATE/20_feature_request.md

@@ -0,0 +1,6 @@
+---
+name: Feature request
+about: Request a new feature
+labels: feature request
+---
+

.github/ISSUE_TEMPLATE/30_model_request.md

@@ -0,0 +1,5 @@
+---
+name: Model request
+about: Request support for a new model to be added to Ollama
+labels: model request
+---

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,8 @@
+blank_issues_enabled: true
+contact_links:
+  - name: Help
+    url: https://discord.com/invite/ollama
+    about: Please join our Discord server for help using Ollama
+  - name: Troubleshooting
+    url: https://github.com/ollama/ollama/blob/main/docs/faq.md#faq
+    about: See the FAQ for common issues and solutions

.github/workflows/latest.yaml

@@ -0,0 +1,24 @@
+name: latest
+
+on:
+  release:
+    types: [released]
+
+jobs:
+  update-latest:
+    environment: release
+    runs-on: linux
+    steps:
+      - uses: actions/checkout@v4
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ vars.DOCKER_USER }}
+          password: ${{ secrets.DOCKER_ACCESS_TOKEN }}
+      - name: Tag images as latest
+        env:
+          PUSH: "1"
+        shell: bash
+        run: |
+          export "VERSION=${GITHUB_REF_NAME#v}"
+          ./scripts/tag_latest.sh

.github/workflows/release.yaml

@@ -0,0 +1,473 @@
+name: release
+
+on:
+  push:
+    tags:
+      - 'v*'
+
+jobs:
+  # Full build of the Mac assets
+  build-darwin:
+    runs-on: macos-12
+    environment: release
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set Version
+        shell: bash
+        run: |
+          echo "VERSION=${GITHUB_REF_NAME#v}" >> $GITHUB_ENV
+          echo "RELEASE_VERSION=$(echo ${GITHUB_REF_NAME} | cut -f1 -d-)" >> $GITHUB_ENV
+      - name: key
+        env:
+          MACOS_SIGNING_KEY: ${{ secrets.MACOS_SIGNING_KEY }}
+          MACOS_SIGNING_KEY_PASSWORD: ${{ secrets.MACOS_SIGNING_KEY_PASSWORD }}
+        run: |
+          echo $MACOS_SIGNING_KEY | base64 --decode > certificate.p12
+          security create-keychain -p password build.keychain
+          security default-keychain -s build.keychain
+          security unlock-keychain -p password build.keychain
+          security import certificate.p12 -k build.keychain -P $MACOS_SIGNING_KEY_PASSWORD -T /usr/bin/codesign
+          security set-key-partition-list -S apple-tool:,apple:,codesign: -s -k password build.keychain
+      - uses: actions/setup-go@v5
+        with:
+          go-version-file: go.mod
+          cache: true
+      - name: Build Darwin
+        env:
+          APPLE_IDENTITY: ${{ secrets.APPLE_IDENTITY }}
+          APPLE_PASSWORD: ${{ secrets.APPLE_PASSWORD }}
+          APPLE_TEAM_ID: ${{ vars.APPLE_TEAM_ID }}
+          APPLE_ID: ${{ vars.APPLE_ID }}
+          SDKROOT: /Applications/Xcode_13.4.1.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk
+          DEVELOPER_DIR: /Applications/Xcode_13.4.1.app/Contents/Developer
+        run: |
+          ./scripts/build_darwin.sh
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist-darwin
+          path: |
+            dist/*arwin*
+            !dist/*-cov
+
+  # Windows builds take a long time to both install the dependencies and build, so parallelize
+  # CPU generation step
+  generate-windows-cpu:
+    environment: release
+    runs-on: windows
+    env:
+      KEY_CONTAINER: ${{ vars.KEY_CONTAINER }}
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set Version
+        shell: bash
+        run: echo "VERSION=${GITHUB_REF_NAME#v}" >> $GITHUB_ENV
+      - uses: 'google-github-actions/auth@v2'
+        with:
+          project_id: 'ollama'
+          credentials_json: '${{ secrets.GOOGLE_SIGNING_CREDENTIALS }}'
+      - run: echo "${{ vars.OLLAMA_CERT }}" > ollama_inc.crt
+      - name: install Windows SDK 8.1 to get signtool
+        run: |
+          $ErrorActionPreference = "Stop"
+          write-host "downloading SDK"
+          Invoke-WebRequest -Uri "https://go.microsoft.com/fwlink/p/?LinkId=323507" -OutFile "${env:RUNNER_TEMP}\sdksetup.exe"
+          Start-Process "${env:RUNNER_TEMP}\sdksetup.exe" -ArgumentList @("/q") -NoNewWindow -Wait
+          write-host "Win SDK 8.1 installed"
+          gci -path 'C:\Program Files (x86)\Windows Kits\' -r -fi 'signtool.exe'
+      - name: install signing plugin
+        run: |
+          $ErrorActionPreference = "Stop"
+          write-host "downloading plugin"
+          Invoke-WebRequest -Uri "https://github.com/GoogleCloudPlatform/kms-integrations/releases/download/cng-v1.0/kmscng-1.0-windows-amd64.zip" -OutFile "${env:RUNNER_TEMP}\plugin.zip"
+          Expand-Archive -Path "${env:RUNNER_TEMP}\plugin.zip" -DestinationPath ${env:RUNNER_TEMP}\plugin\
+          write-host "Installing plugin"
+          & "${env:RUNNER_TEMP}\plugin\*\kmscng.msi" /quiet
+          write-host "plugin installed"
+      - uses: actions/setup-go@v5
+        with:
+          go-version-file: go.mod
+          cache: true
+      - run: go get ./...
+      - run: |
+          $gopath=(get-command go).source | split-path -parent
+          & "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\Common7\Tools\Launch-VsDevShell.ps1"
+          cd $env:GITHUB_WORKSPACE
+          $env:CMAKE_SYSTEM_VERSION="10.0.22621.0"
+          $env:PATH="$gopath;$env:PATH"
+          go generate -x ./...
+        name: go generate
+      - uses: actions/upload-artifact@v4
+        with:
+          name: generate-windows-cpu
+          path: |
+            llm/build/**/bin/*
+            llm/build/**/*.a
+            dist/windows-amd64/**
+
+  # ROCm generation step
+  generate-windows-rocm:
+    environment: release
+    runs-on: windows
+    env:
+      KEY_CONTAINER: ${{ vars.KEY_CONTAINER }}
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set Version
+        shell: bash
+        run: echo "VERSION=${GITHUB_REF_NAME#v}" >> $GITHUB_ENV
+      - uses: 'google-github-actions/auth@v2'
+        with:
+          project_id: 'ollama'
+          credentials_json: '${{ secrets.GOOGLE_SIGNING_CREDENTIALS }}'
+      - run: echo "${{ vars.OLLAMA_CERT }}" > ollama_inc.crt
+      - name: install Windows SDK 8.1 to get signtool
+        run: |
+          $ErrorActionPreference = "Stop"
+          write-host "downloading SDK"
+          Invoke-WebRequest -Uri "https://go.microsoft.com/fwlink/p/?LinkId=323507" -OutFile "${env:RUNNER_TEMP}\sdksetup.exe"
+          Start-Process "${env:RUNNER_TEMP}\sdksetup.exe" -ArgumentList @("/q") -NoNewWindow -Wait
+          write-host "Win SDK 8.1 installed"
+          gci -path 'C:\Program Files (x86)\Windows Kits\' -r -fi 'signtool.exe'
+      - name: install signing plugin
+        run: |
+          $ErrorActionPreference = "Stop"
+          write-host "downloading plugin"
+          Invoke-WebRequest -Uri "https://github.com/GoogleCloudPlatform/kms-integrations/releases/download/cng-v1.0/kmscng-1.0-windows-amd64.zip" -OutFile "${env:RUNNER_TEMP}\plugin.zip"
+          Expand-Archive -Path "${env:RUNNER_TEMP}\plugin.zip" -DestinationPath ${env:RUNNER_TEMP}\plugin\
+          write-host "Installing plugin"
+          & "${env:RUNNER_TEMP}\plugin\*\kmscng.msi" /quiet
+          write-host "plugin installed"
+      - uses: actions/setup-go@v5
+        with:
+          go-version-file: go.mod
+          cache: true
+      - name: 'Install ROCm'
+        run: |
+          $ErrorActionPreference = "Stop"
+          write-host "downloading AMD HIP Installer"
+          Invoke-WebRequest -Uri "https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-23.Q4-WinSvr2022-For-HIP.exe" -OutFile "${env:RUNNER_TEMP}\rocm-install.exe"
+          write-host "Installing AMD HIP"
+          Start-Process "${env:RUNNER_TEMP}\rocm-install.exe" -ArgumentList '-install' -NoNewWindow -Wait
+          write-host "Completed AMD HIP"
+      - name: 'Verify ROCm'
+        run: |
+          & 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' --version
+      - run: go get ./...
+      - run: |
+          $gopath=(get-command go).source | split-path -parent
+          & "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\Common7\Tools\Launch-VsDevShell.ps1"
+          cd $env:GITHUB_WORKSPACE
+          $env:CMAKE_SYSTEM_VERSION="10.0.22621.0"
+          $env:PATH="$gopath;$env:PATH"
+          $env:OLLAMA_SKIP_CPU_GENERATE="1"
+          $env:HIP_PATH=$(Resolve-Path 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | split-path | split-path)
+          go generate -x ./...
+        name: go generate
+      - name: 'gather rocm dependencies'
+        run: |
+          $HIP_PATH=$(Resolve-Path 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | split-path | split-path)
+          md "dist\deps\bin\rocblas\library"
+          cp "${HIP_PATH}\bin\hipblas.dll" "dist\deps\bin\"
+          cp "${HIP_PATH}\bin\rocblas.dll" "dist\deps\bin\"
+          cp "${HIP_PATH}\bin\rocblas\library\*" "dist\deps\bin\rocblas\library\"
+      - uses: actions/upload-artifact@v4
+        with:
+          name: generate-windows-rocm
+          path: |
+            llm/build/**/bin/*
+            dist/windows-amd64/**
+      - uses: actions/upload-artifact@v4
+        with:
+          name: windows-rocm-deps
+          path: dist/deps/*
+
+  # CUDA generation step
+  generate-windows-cuda:
+    environment: release
+    runs-on: windows
+    env:
+      KEY_CONTAINER: ${{ vars.KEY_CONTAINER }}
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set Version
+        shell: bash
+        run: echo "VERSION=${GITHUB_REF_NAME#v}" >> $GITHUB_ENV
+      - uses: 'google-github-actions/auth@v2'
+        with:
+          project_id: 'ollama'
+          credentials_json: '${{ secrets.GOOGLE_SIGNING_CREDENTIALS }}'
+      - run: echo "${{ vars.OLLAMA_CERT }}" > ollama_inc.crt
+      - name: install Windows SDK 8.1 to get signtool
+        run: |
+          $ErrorActionPreference = "Stop"
+          write-host "downloading SDK"
+          Invoke-WebRequest -Uri "https://go.microsoft.com/fwlink/p/?LinkId=323507" -OutFile "${env:RUNNER_TEMP}\sdksetup.exe"
+          Start-Process "${env:RUNNER_TEMP}\sdksetup.exe" -ArgumentList @("/q") -NoNewWindow -Wait
+          write-host "Win SDK 8.1 installed"
+          gci -path 'C:\Program Files (x86)\Windows Kits\' -r -fi 'signtool.exe'
+      - name: install signing plugin
+        run: |
+          $ErrorActionPreference = "Stop"
+          write-host "downloading plugin"
+          Invoke-WebRequest -Uri "https://github.com/GoogleCloudPlatform/kms-integrations/releases/download/cng-v1.0/kmscng-1.0-windows-amd64.zip" -OutFile "${env:RUNNER_TEMP}\plugin.zip"
+          Expand-Archive -Path "${env:RUNNER_TEMP}\plugin.zip" -DestinationPath ${env:RUNNER_TEMP}\plugin\
+          write-host "Installing plugin"
+          & "${env:RUNNER_TEMP}\plugin\*\kmscng.msi" /quiet
+          write-host "plugin installed"
+      - uses: actions/setup-go@v5
+        with:
+          go-version-file: go.mod
+          cache: true
+      - name: 'Install CUDA'
+        run: |
+          $ErrorActionPreference = "Stop"
+          write-host "downloading CUDA Installer"
+          Invoke-WebRequest -Uri "https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.89_win10.exe" -OutFile "${env:RUNNER_TEMP}\cuda-install.exe"
+          write-host "Installing CUDA"
+          Start-Process "${env:RUNNER_TEMP}\cuda-install.exe" -ArgumentList '-s' -NoNewWindow -Wait
+          write-host "Completed CUDA"
+          $cudaPath=((resolve-path "c:\Program Files\NVIDIA*\CUDA\v*\bin\nvcc.exe")[0].path | split-path | split-path)
+          $cudaVer=($cudaPath | split-path -leaf ) -replace 'v(\d+).(\d+)', '$1_$2' 
+          echo "$cudaPath\bin" >> $env:GITHUB_PATH
+          echo "CUDA_PATH=$cudaPath" >> $env:GITHUB_ENV
+          echo "CUDA_PATH_V${cudaVer}=$cudaPath" >> $env:GITHUB_ENV
+          echo "CUDA_PATH_VX_Y=CUDA_PATH_V${cudaVer}" >> $env:GITHUB_ENV
+      - name: 'Verify CUDA'
+        run: nvcc -V
+      - run: go get ./...
+      - name: go generate
+        run: |
+          $gopath=(get-command go).source | split-path -parent
+          $cudabin=(get-command nvcc).source | split-path
+          & "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\Common7\Tools\Launch-VsDevShell.ps1"
+          cd $env:GITHUB_WORKSPACE
+          $env:CMAKE_SYSTEM_VERSION="10.0.22621.0"
+          $env:PATH="$gopath;$cudabin;$env:PATH"
+          $env:OLLAMA_SKIP_CPU_GENERATE="1"
+          go generate -x ./...
+      - name: 'gather cuda dependencies'
+        run: |
+          $NVIDIA_DIR=(resolve-path 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\*\bin\')[0]
+          md "dist\deps"
+          cp "${NVIDIA_DIR}\cudart64_*.dll" "dist\deps\"
+          cp "${NVIDIA_DIR}\cublas64_*.dll" "dist\deps\"
+          cp "${NVIDIA_DIR}\cublasLt64_*.dll" "dist\deps\"
+      - uses: actions/upload-artifact@v4
+        with:
+          name: generate-windows-cuda
+          path: |
+            llm/build/**/bin/*
+            dist/windows-amd64/**
+      - uses: actions/upload-artifact@v4
+        with:
+          name: windows-cuda-deps
+          path: dist/deps/*
+
+  # Import the prior generation steps and build the final windows assets
+  build-windows:
+    environment: release
+    runs-on: windows
+    needs:
+      - generate-windows-cuda
+      - generate-windows-rocm
+      - generate-windows-cpu
+    env:
+      KEY_CONTAINER: ${{ vars.KEY_CONTAINER }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: recursive
+      - name: Set Version
+        shell: bash
+        run: echo "VERSION=${GITHUB_REF_NAME#v}" >> $GITHUB_ENV
+      - uses: 'google-github-actions/auth@v2'
+        with:
+          project_id: 'ollama'
+          credentials_json: '${{ secrets.GOOGLE_SIGNING_CREDENTIALS }}'
+      - run: echo "${{ vars.OLLAMA_CERT }}" > ollama_inc.crt
+      - name: install Windows SDK 8.1 to get signtool
+        run: |
+          $ErrorActionPreference = "Stop"
+          write-host "downloading SDK"
+          Invoke-WebRequest -Uri "https://go.microsoft.com/fwlink/p/?LinkId=323507" -OutFile "${env:RUNNER_TEMP}\sdksetup.exe"
+          Start-Process "${env:RUNNER_TEMP}\sdksetup.exe" -ArgumentList @("/q") -NoNewWindow -Wait
+          write-host "Win SDK 8.1 installed"
+          gci -path 'C:\Program Files (x86)\Windows Kits\' -r -fi 'signtool.exe'
+      - name: install signing plugin
+        run: |
+          $ErrorActionPreference = "Stop"
+          write-host "downloading plugin"
+          Invoke-WebRequest -Uri "https://github.com/GoogleCloudPlatform/kms-integrations/releases/download/cng-v1.0/kmscng-1.0-windows-amd64.zip" -OutFile "${env:RUNNER_TEMP}\plugin.zip"
+          Expand-Archive -Path "${env:RUNNER_TEMP}\plugin.zip" -DestinationPath ${env:RUNNER_TEMP}\plugin\
+          write-host "Installing plugin"
+          & "${env:RUNNER_TEMP}\plugin\*\kmscng.msi" /quiet
+          write-host "plugin installed"
+      - uses: actions/setup-go@v5
+        with:
+          go-version-file: go.mod
+          cache: true
+      - run: go get
+      - uses: actions/download-artifact@v4
+        with:
+          name: generate-windows-cpu
+      - uses: actions/download-artifact@v4
+        with:
+          name: generate-windows-cuda
+      - uses: actions/download-artifact@v4
+        with:
+          name: windows-cuda-deps
+      - uses: actions/download-artifact@v4
+        with:
+          name: windows-rocm-deps
+      - uses: actions/download-artifact@v4
+        with:
+          name: generate-windows-rocm
+      - run: dir llm/build
+      - run: |
+          $gopath=(get-command go).source | split-path -parent
+          & "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\Common7\Tools\Launch-VsDevShell.ps1"
+          cd $env:GITHUB_WORKSPACE
+          $env:CMAKE_SYSTEM_VERSION="10.0.22621.0"
+          $env:PATH="$gopath;$env:PATH"
+          $env:OLLAMA_SKIP_GENERATE="1"
+          & .\scripts\build_windows.ps1
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist-windows
+          path: |
+            dist/OllamaSetup.exe
+            dist/ollama-windows-*.zip
+
+  # Linux x86 assets built using the container based build
+  build-linux-amd64:
+    environment: release
+    runs-on: linux
+    env:
+      OLLAMA_SKIP_MANIFEST_CREATE: '1'
+      BUILD_ARCH: amd64
+      PUSH: '1'
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: recursive
+      - name: Set Version
+        shell: bash
+        run: echo "VERSION=${GITHUB_REF_NAME#v}" >> $GITHUB_ENV
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ vars.DOCKER_USER }}
+          password: ${{ secrets.DOCKER_ACCESS_TOKEN }}
+      - run: |
+          ./scripts/build_linux.sh
+          ./scripts/build_docker.sh
+          mv dist/deps/* dist/
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist-linux-amd64
+          path: |
+            dist/*linux*
+            !dist/*-cov
+
+  # Linux ARM assets built using the container based build
+  # (at present, docker isn't pre-installed on arm ubunutu images)
+  build-linux-arm64:
+    environment: release
+    runs-on: linux-arm64
+    env:
+      OLLAMA_SKIP_MANIFEST_CREATE: '1'
+      BUILD_ARCH: arm64
+      PUSH: '1'
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: recursive
+      - name: Set Version
+        shell: bash
+        run: echo "VERSION=${GITHUB_REF_NAME#v}" >> $GITHUB_ENV
+      - name: 'Install Docker'
+        run: |
+          # Add Docker's official GPG key:
+          env
+          uname -a
+          sudo apt-get update
+          sudo apt-get install -y ca-certificates curl
+          sudo install -m 0755 -d /etc/apt/keyrings
+          sudo curl -fsSL https://download.docker.com/linux/ubuntu/gpg -o /etc/apt/keyrings/docker.asc
+          sudo chmod a+r /etc/apt/keyrings/docker.asc
+
+          # Add the repository to Apt sources:
+          echo \
+            "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/docker.asc] https://download.docker.com/linux/ubuntu \
+            $(. /etc/os-release && echo "$VERSION_CODENAME") stable" | \
+            sudo tee /etc/apt/sources.list.d/docker.list > /dev/null
+          sudo apt-get update
+          sudo apt-get install -y docker-ce docker-ce-cli containerd.io
+          sudo usermod -aG docker $USER
+          sudo apt-get install acl
+          sudo setfacl --modify user:$USER:rw /var/run/docker.sock
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ vars.DOCKER_USER }}
+          password: ${{ secrets.DOCKER_ACCESS_TOKEN }}
+      - run: |
+          ./scripts/build_linux.sh
+          ./scripts/build_docker.sh
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist-linux-arm64
+          path: |
+            dist/*linux*
+            !dist/*-cov
+
+  # Aggregate all the assets and ship a release
+  release:
+    needs:
+      - build-darwin
+      - build-windows
+      - build-linux-amd64
+      - build-linux-arm64
+    runs-on: linux
+    environment: release
+    permissions:
+      contents: write
+    env:
+      OLLAMA_SKIP_IMAGE_BUILD: '1'
+      PUSH: '1'
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set Version
+        shell: bash
+        run: |
+          echo "VERSION=${GITHUB_REF_NAME#v}" >> $GITHUB_ENV
+          echo "RELEASE_VERSION=$(echo ${GITHUB_REF_NAME} | cut -f1 -d-)" >> $GITHUB_ENV
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ vars.DOCKER_USER }}
+          password: ${{ secrets.DOCKER_ACCESS_TOKEN }}
+      - run: ./scripts/build_docker.sh
+      - name: Retrieve built artifact
+        uses: actions/download-artifact@v4
+        with:
+          path: dist
+          pattern: dist-*
+          merge-multiple: true
+      - run: |
+          ls -lh dist/
+          (cd dist; sha256sum * > sha256sum.txt)
+          cat dist/sha256sum.txt
+      - uses: ncipollo/release-action@v1
+        with:
+          name: ${{ env.RELEASE_VERSION }}
+          allowUpdates: true
+          artifacts: 'dist/*'
+          draft: true
+          prerelease: true
+          omitBodyDuringUpdate: true
+          generateReleaseNotes: true
+          omitDraftDuringUpdate: true
+          omitPrereleaseDuringUpdate: true
+          replacesArtifacts: true

.github/workflows/test.yaml

@@ -1,18 +1,59 @@
 name: test
 
+concurrency:
+  # For PRs, later CI runs preempt previous ones. e.g. a force push on a PR
+  # cancels running CI jobs and starts all new ones.
+  #
+  # For non-PR pushes, concurrency.group needs to be unique for every distinct
+  # CI run we want to have happen. Use run_id, which in practice means all
+  # non-PR CI runs will be allowed to run without preempting each other.
+  group: ${{ github.workflow }}-$${{ github.pull_request.number || github.run_id }}
+  cancel-in-progress: true
+
 on:
   pull_request:
+    paths:
+      - '**/*'
+      - '!docs/**'
+      - '!README.md'
 
 jobs:
+  changes:
+    runs-on: ubuntu-latest
+    outputs:
+      GENERATE: ${{ steps.changes.outputs.GENERATE }}
+      GENERATE_CUDA: ${{ steps.changes.outputs.GENERATE_CUDA }}
+      GENERATE_ROCM: ${{ steps.changes.outputs.GENERATE_ROCM }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+      - id: changes
+        run: |
+          changed() {
+            git diff-tree -r --no-commit-id --name-only \
+              $(git merge-base ${{ github.event.pull_request.base.sha }} ${{ github.event.pull_request.head.sha }}) \
+              ${{ github.event.pull_request.head.sha }} \
+              | xargs python3 -c "import sys; print(any([x.startswith('$1') for x in sys.argv[1:]]))"
+          }
+
+          {
+            echo GENERATE=$(changed llm/)
+            echo GENERATE_CUDA=$(changed llm/)
+            echo GENERATE_ROCM=$(changed llm/)
+          } >>$GITHUB_OUTPUT
+
   generate:
+    needs: [changes]
+    if: ${{ needs.changes.outputs.GENERATE == 'True' }}
     strategy:
       matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
+        os: [ubuntu-latest, macos-latest, windows-2019]
         arch: [amd64, arm64]
         exclude:
           - os: ubuntu-latest
             arch: arm64
-          - os: windows-latest
+          - os: windows-2019
             arch: arm64
     runs-on: ${{ matrix.os }}
     env:
@@ -21,86 +62,258 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions/setup-go@v5
         with:
-          go-version: '1.21'
+          go-version-file: go.mod
           cache: true
-      - if: ${{ startsWith(matrix.os, 'windows-') }}
-        shell: pwsh
-        run: |
-          $path = vswhere -latest -products * -requires Microsoft.VisualStudio.Component.VC.Tools.x86.x64 -property installationPath
-          if ($path) {
-              $path = join-path $path 'Common7\Tools\vsdevcmd.bat'
-              if (test-path $path) {
-                  cmd /s /c """$path"" $args && set" | where { $_ -match '(\w+)=(.*)' } | foreach {
-                      echo "$($Matches[1])=$($Matches[2])" | Out-File -FilePath $Env:GITHUB_ENV -Encoding utf8 -Append
-                  }
-              }
-          }
-
-          echo "C:\Program Files\Git\usr\bin" | Out-File -FilePath $Env:GITHUB_PATH -Encoding utf8 -Append
       - run: go get ./...
+      - run: |
+          $gopath=(get-command go).source | split-path -parent
+          $gccpath=(get-command gcc).source | split-path -parent
+          & "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\Common7\Tools\Launch-VsDevShell.ps1"
+          cd $env:GITHUB_WORKSPACE
+          $env:CMAKE_SYSTEM_VERSION="10.0.22621.0"
+          $env:PATH="$gopath;$gccpath;$env:PATH"
+          echo $env:PATH
+          go generate -x ./...
+        if: ${{ startsWith(matrix.os, 'windows-') }}
+        name: 'Windows Go Generate'
       - run: go generate -x ./...
+        if: ${{ ! startsWith(matrix.os, 'windows-') }}
+        name: 'Unix Go Generate'
       - uses: actions/upload-artifact@v4
         with:
           name: ${{ matrix.os }}-${{ matrix.arch }}-libraries
           path: |
-            llm/llama.cpp/build/**/lib/*
+            llm/build/**/bin/*
-  lint:
+            llm/build/**/*.a
-    needs: generate
+  generate-cuda:
+    needs: [changes]
+    if: ${{ needs.changes.outputs.GENERATE_CUDA == 'True' }}
     strategy:
       matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
+        cuda-version:
+          - '11.8.0'
+    runs-on: linux
+    container: nvidia/cuda:${{ matrix.cuda-version }}-devel-ubuntu20.04
+    steps:
+      - run: |
+          apt-get update && apt-get install -y git build-essential curl
+          curl -fsSL https://github.com/Kitware/CMake/releases/download/v3.28.1/cmake-3.28.1-linux-x86_64.tar.gz \
+            | tar -zx -C /usr --strip-components 1
+        env:
+          DEBIAN_FRONTEND: noninteractive
+      - uses: actions/checkout@v4
+      - uses: actions/setup-go@v4
+        with:
+          go-version-file: go.mod
+          cache: true
+      - run: go get ./...
+      - run: |
+          git config --global --add safe.directory /__w/ollama/ollama
+          go generate -x ./...
+        env:
+          OLLAMA_SKIP_CPU_GENERATE: '1'
+      - uses: actions/upload-artifact@v4
+        with:
+          name: cuda-${{ matrix.cuda-version }}-libraries
+          path: |
+            llm/build/**/bin/*
+            dist/windows-amd64/**
+  generate-rocm:
+    needs: [changes]
+    if: ${{ needs.changes.outputs.GENERATE_ROCM == 'True' }}
+    strategy:
+      matrix:
+        rocm-version:
+          - '6.0.2'
+    runs-on: linux
+    container: rocm/dev-ubuntu-20.04:${{ matrix.rocm-version }}
+    steps:
+      - run: |
+          apt-get update && apt-get install -y git build-essential curl rocm-libs
+          curl -fsSL https://github.com/Kitware/CMake/releases/download/v3.28.1/cmake-3.28.1-linux-x86_64.tar.gz \
+            | tar -zx -C /usr --strip-components 1
+        env:
+          DEBIAN_FRONTEND: noninteractive
+      - uses: actions/checkout@v4
+      - uses: actions/setup-go@v4
+        with:
+          go-version-file: go.mod
+          cache: true
+      - run: go get ./...
+      - run: |
+          git config --global --add safe.directory /__w/ollama/ollama
+          go generate -x ./...
+        env:
+          OLLAMA_SKIP_CPU_GENERATE: '1'
+      - uses: actions/upload-artifact@v4
+        with:
+          name: rocm-${{ matrix.rocm-version }}-libraries
+          path: |
+            llm/build/**/bin/*
+            dist/windows-amd64/**
+
+  # ROCm generation step
+  generate-windows-rocm:
+    needs: [changes]
+    if: ${{ needs.changes.outputs.GENERATE_ROCM == 'True' }}
+    runs-on: windows
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-go@v5
+        with:
+          go-version-file: go.mod
+          cache: true
+      - name: 'Install ROCm'
+        run: |
+          $ErrorActionPreference = "Stop"
+          write-host "downloading AMD HIP Installer"
+          Invoke-WebRequest -Uri "https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-23.Q4-WinSvr2022-For-HIP.exe" -OutFile "${env:RUNNER_TEMP}\rocm-install.exe"
+          write-host "Installing AMD HIP"
+          Start-Process "${env:RUNNER_TEMP}\rocm-install.exe" -ArgumentList '-install' -NoNewWindow -Wait
+          write-host "Completed AMD HIP"
+      - name: 'Verify ROCm'
+        run: |
+          & 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' --version
+      - run: go get ./...
+      - run: |
+          $gopath=(get-command go).source | split-path -parent
+          & "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\Common7\Tools\Launch-VsDevShell.ps1"
+          cd $env:GITHUB_WORKSPACE
+          $env:CMAKE_SYSTEM_VERSION="10.0.22621.0"
+          $env:PATH="$gopath;$env:PATH"
+          $env:OLLAMA_SKIP_CPU_GENERATE="1"
+          $env:HIP_PATH=$(Resolve-Path 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | split-path | split-path)
+          go generate -x ./...
+        name: go generate
+        env:
+          OLLAMA_SKIP_CPU_GENERATE: '1'
+      # TODO - do we need any artifacts?
+
+  # CUDA generation step
+  generate-windows-cuda:
+    needs: [changes]
+    if: ${{ needs.changes.outputs.GENERATE_CUDA == 'True' }}
+    runs-on: windows
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-go@v5
+        with:
+          go-version-file: go.mod
+          cache: true
+      - name: 'Install CUDA'
+        run: |
+          $ErrorActionPreference = "Stop"
+          write-host "downloading CUDA Installer"
+          Invoke-WebRequest -Uri "https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.89_win10.exe" -OutFile "${env:RUNNER_TEMP}\cuda-install.exe"
+          write-host "Installing CUDA"
+          Start-Process "${env:RUNNER_TEMP}\cuda-install.exe" -ArgumentList '-s' -NoNewWindow -Wait
+          write-host "Completed CUDA"
+          $cudaPath=((resolve-path "c:\Program Files\NVIDIA*\CUDA\v*\bin\nvcc.exe")[0].path | split-path | split-path)
+          $cudaVer=($cudaPath | split-path -leaf ) -replace 'v(\d+).(\d+)', '$1_$2' 
+          echo "$cudaPath\bin" >> $env:GITHUB_PATH
+          echo "CUDA_PATH=$cudaPath" >> $env:GITHUB_ENV
+          echo "CUDA_PATH_V${cudaVer}=$cudaPath" >> $env:GITHUB_ENV
+          echo "CUDA_PATH_VX_Y=CUDA_PATH_V${cudaVer}" >> $env:GITHUB_ENV
+      - name: 'Verify CUDA'
+        run: nvcc -V
+      - run: go get ./...
+      - name: go generate
+        run: |
+          $gopath=(get-command go).source | split-path -parent
+          $cudabin=(get-command nvcc).source | split-path
+          & "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\Common7\Tools\Launch-VsDevShell.ps1"
+          cd $env:GITHUB_WORKSPACE
+          $env:CMAKE_SYSTEM_VERSION="10.0.22621.0"
+          $env:PATH="$gopath;$cudabin;$env:PATH"
+          $env:OLLAMA_SKIP_CPU_GENERATE="1"
+          go generate -x ./...
+        env:
+          OLLAMA_SKIP_CPU_GENERATE: '1'
+      # TODO - do we need any artifacts?
+
+  lint:
+    strategy:
+      matrix:
+        os: [ubuntu-latest, macos-latest, windows-2019]
         arch: [amd64, arm64]
         exclude:
           - os: ubuntu-latest
             arch: arm64
-          - os: windows-latest
+          - os: windows-2019
             arch: arm64
           - os: macos-latest
             arch: amd64
     runs-on: ${{ matrix.os }}
     env:
       GOARCH: ${{ matrix.arch }}
-      CGO_ENABLED: "1"
+      CGO_ENABLED: '1'
     steps:
       - uses: actions/checkout@v4
         with:
           submodules: recursive
       - uses: actions/setup-go@v5
         with:
-          go-version: '1.21'
+          go-version-file: go.mod
           cache: false
-      - uses: actions/download-artifact@v4
+      - run: |
+          case ${{ matrix.arch }} in
+            amd64) echo ARCH=x86_64 ;;
+            arm64) echo ARCH=arm64 ;;
+          esac >>$GITHUB_ENV
+        shell: bash
+      - run: |
+          mkdir -p llm/build/linux/$ARCH/stub/bin
+          touch llm/build/linux/$ARCH/stub/bin/ollama_llama_server
+        if: ${{ startsWith(matrix.os, 'ubuntu-') }}
+      - run: |
+          mkdir -p llm/build/darwin/$ARCH/stub/bin
+          touch llm/build/darwin/$ARCH/stub/bin/ollama_llama_server
+        if: ${{ startsWith(matrix.os, 'macos-') }}
+      - uses: golangci/golangci-lint-action@v4
         with:
-          name: ${{ matrix.os }}-${{ matrix.arch }}-libraries
+          args: --timeout 8m0s -v
-          path: llm/llama.cpp/build
-      - uses: golangci/golangci-lint-action@v3
   test:
-    needs: generate
     strategy:
       matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
+        os: [ubuntu-latest, macos-latest, windows-2019]
         arch: [amd64]
         exclude:
           - os: ubuntu-latest
             arch: arm64
-          - os: windows-latest
+          - os: windows-2019
             arch: arm64
     runs-on: ${{ matrix.os }}
     env:
       GOARCH: ${{ matrix.arch }}
-      CGO_ENABLED: "1"
+      CGO_ENABLED: '1'
+      OLLAMA_CPU_TARGET: 'static'
     steps:
       - uses: actions/checkout@v4
         with:
           submodules: recursive
       - uses: actions/setup-go@v5
         with:
-          go-version: '1.21'
+          go-version-file: go.mod
           cache: true
-      - run: go get
+      - run: |
-      - uses: actions/download-artifact@v4
+          case ${{ matrix.arch }} in
-        with:
+            amd64) echo ARCH=x86_64 ;;
-          name: ${{ matrix.os }}-${{ matrix.arch }}-libraries
+            arm64) echo ARCH=arm64 ;;
-          path: llm/llama.cpp/build
+          esac >>$GITHUB_ENV
+        shell: bash
+      - run: |
+          mkdir -p llm/build/linux/$ARCH/stub/bin
+          touch llm/build/linux/$ARCH/stub/bin/ollama_llama_server
+        if: ${{ startsWith(matrix.os, 'ubuntu-') }}
+      - run: |
+          mkdir -p llm/build/darwin/$ARCH/stub/bin
+          touch llm/build/darwin/$ARCH/stub/bin/ollama_llama_server
+        if: ${{ startsWith(matrix.os, 'macos-') }}
+        shell: bash
+      - run: go generate ./...
       - run: go build
       - run: go test -v ./...
+      - uses: actions/upload-artifact@v4
+        with:
+          name: ${{ matrix.os }}-binaries
+          path: ollama

.gitignore

@@ -9,4 +9,6 @@ ggml-metal.metal
 .cache
 *.exe
 .idea
-test_data
+test_data
+*.crt
+llm/build

.golangci.yaml

@@ -15,13 +15,3 @@ linters:
     - misspell
     - nilerr
     - unused
-linters-settings:
-  errcheck:
-    # exclude the following functions since we don't generally
-    # need to be concerned with the returned errors
-    exclude-functions:
-      - encoding/binary.Read
-      - (*os.File).Seek
-      - (*bufio.Writer).WriteString
-      - (*github.com/spf13/pflag.FlagSet).Set
-      - (*github.com/jmorganca/ollama/llm.readSeekOffset).Seek

Dockerfile

@@ -1,29 +1,144 @@
-FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
+ARG GOLANG_VERSION=1.22.1
+ARG CMAKE_VERSION=3.22.1
+# this CUDA_VERSION corresponds with the one specified in docs/gpu.md
+ARG CUDA_VERSION=11.3.1
+ARG ROCM_VERSION=6.0.2
 
-ARG TARGETARCH
+# Copy the minimal context we need to run the generate scripts
-ARG GOFLAGS="'-ldflags=-w -s'"
+FROM scratch AS llm-code
+COPY .git .git
+COPY .gitmodules .gitmodules
+COPY llm llm
 
-WORKDIR /go/src/github.com/jmorganca/ollama
+FROM --platform=linux/amd64 nvidia/cuda:$CUDA_VERSION-devel-centos7 AS cuda-build-amd64
-RUN apt-get update && apt-get install -y git build-essential cmake
+ARG CMAKE_VERSION
-ADD https://dl.google.com/go/go1.21.3.linux-$TARGETARCH.tar.gz /tmp/go1.21.3.tar.gz
+COPY ./scripts/rh_linux_deps.sh /
-RUN mkdir -p /usr/local && tar xz -C /usr/local </tmp/go1.21.3.tar.gz
+RUN CMAKE_VERSION=${CMAKE_VERSION} sh /rh_linux_deps.sh
+ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
+COPY --from=llm-code / /go/src/github.com/ollama/ollama/
+WORKDIR /go/src/github.com/ollama/ollama/llm/generate
+ARG CGO_CFLAGS
+RUN OLLAMA_SKIP_STATIC_GENERATE=1 OLLAMA_SKIP_CPU_GENERATE=1 sh gen_linux.sh
 
+FROM --platform=linux/arm64 nvidia/cuda:$CUDA_VERSION-devel-rockylinux8 AS cuda-build-arm64
+ARG CMAKE_VERSION
+COPY ./scripts/rh_linux_deps.sh /
+RUN CMAKE_VERSION=${CMAKE_VERSION} sh /rh_linux_deps.sh
+ENV PATH /opt/rh/gcc-toolset-10/root/usr/bin:$PATH
+COPY --from=llm-code / /go/src/github.com/ollama/ollama/
+WORKDIR /go/src/github.com/ollama/ollama/llm/generate
+ARG CGO_CFLAGS
+RUN OLLAMA_SKIP_STATIC_GENERATE=1 OLLAMA_SKIP_CPU_GENERATE=1 sh gen_linux.sh
+
+FROM --platform=linux/amd64 rocm/dev-centos-7:${ROCM_VERSION}-complete AS rocm-build-amd64
+ARG CMAKE_VERSION
+COPY ./scripts/rh_linux_deps.sh /
+RUN CMAKE_VERSION=${CMAKE_VERSION} sh /rh_linux_deps.sh
+ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
+ENV LIBRARY_PATH /opt/amdgpu/lib64
+COPY --from=llm-code / /go/src/github.com/ollama/ollama/
+WORKDIR /go/src/github.com/ollama/ollama/llm/generate
+ARG CGO_CFLAGS
+ARG AMDGPU_TARGETS
+RUN OLLAMA_SKIP_STATIC_GENERATE=1 OLLAMA_SKIP_CPU_GENERATE=1 sh gen_linux.sh
+RUN mkdir /tmp/scratch && \
+    for dep in $(zcat /go/src/github.com/ollama/ollama/llm/build/linux/x86_64/rocm*/bin/deps.txt.gz) ; do \
+        cp ${dep} /tmp/scratch/ || exit 1 ; \
+    done && \
+    (cd /opt/rocm/lib && tar cf - rocblas/library) | (cd /tmp/scratch/ && tar xf - ) && \
+    mkdir -p /go/src/github.com/ollama/ollama/dist/deps/ && \
+    (cd /tmp/scratch/ && tar czvf /go/src/github.com/ollama/ollama/dist/deps/ollama-linux-amd64-rocm.tgz . )
+
+
+FROM --platform=linux/amd64 centos:7 AS cpu-builder-amd64
+ARG CMAKE_VERSION
+ARG GOLANG_VERSION
+COPY ./scripts/rh_linux_deps.sh /
+RUN CMAKE_VERSION=${CMAKE_VERSION} GOLANG_VERSION=${GOLANG_VERSION} sh /rh_linux_deps.sh
+ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
+COPY --from=llm-code / /go/src/github.com/ollama/ollama/
+ARG OLLAMA_CUSTOM_CPU_DEFS
+ARG CGO_CFLAGS
+WORKDIR /go/src/github.com/ollama/ollama/llm/generate
+
+FROM --platform=linux/amd64 cpu-builder-amd64 AS static-build-amd64
+RUN OLLAMA_CPU_TARGET="static" sh gen_linux.sh
+FROM --platform=linux/amd64 cpu-builder-amd64 AS cpu-build-amd64
+RUN OLLAMA_SKIP_STATIC_GENERATE=1 OLLAMA_CPU_TARGET="cpu" sh gen_linux.sh
+FROM --platform=linux/amd64 cpu-builder-amd64 AS cpu_avx-build-amd64
+RUN OLLAMA_SKIP_STATIC_GENERATE=1 OLLAMA_CPU_TARGET="cpu_avx" sh gen_linux.sh
+FROM --platform=linux/amd64 cpu-builder-amd64 AS cpu_avx2-build-amd64
+RUN OLLAMA_SKIP_STATIC_GENERATE=1 OLLAMA_CPU_TARGET="cpu_avx2" sh gen_linux.sh
+
+FROM --platform=linux/arm64 centos:7 AS cpu-builder-arm64
+ARG CMAKE_VERSION
+ARG GOLANG_VERSION
+COPY ./scripts/rh_linux_deps.sh /
+RUN CMAKE_VERSION=${CMAKE_VERSION} GOLANG_VERSION=${GOLANG_VERSION} sh /rh_linux_deps.sh
+ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
+COPY --from=llm-code / /go/src/github.com/ollama/ollama/
+ARG OLLAMA_CUSTOM_CPU_DEFS
+ARG CGO_CFLAGS
+WORKDIR /go/src/github.com/ollama/ollama/llm/generate
+
+FROM --platform=linux/arm64 cpu-builder-arm64 AS static-build-arm64
+RUN OLLAMA_CPU_TARGET="static" sh gen_linux.sh
+FROM --platform=linux/arm64 cpu-builder-arm64 AS cpu-build-arm64
+RUN OLLAMA_SKIP_STATIC_GENERATE=1 OLLAMA_CPU_TARGET="cpu" sh gen_linux.sh
+
+
+# Intermediate stage used for ./scripts/build_linux.sh
+FROM --platform=linux/amd64 cpu-build-amd64 AS build-amd64
+ENV CGO_ENABLED 1
+WORKDIR /go/src/github.com/ollama/ollama
 COPY . .
-ENV GOARCH=$TARGETARCH
+COPY --from=static-build-amd64 /go/src/github.com/ollama/ollama/llm/build/linux/ llm/build/linux/
-ENV GOFLAGS=$GOFLAGS
+COPY --from=cpu_avx-build-amd64 /go/src/github.com/ollama/ollama/llm/build/linux/ llm/build/linux/
-RUN /usr/local/go/bin/go generate ./... \
+COPY --from=cpu_avx2-build-amd64 /go/src/github.com/ollama/ollama/llm/build/linux/ llm/build/linux/
-    && /usr/local/go/bin/go build .
+COPY --from=cuda-build-amd64 /go/src/github.com/ollama/ollama/llm/build/linux/ llm/build/linux/
+COPY --from=rocm-build-amd64 /go/src/github.com/ollama/ollama/llm/build/linux/ llm/build/linux/
+COPY --from=rocm-build-amd64 /go/src/github.com/ollama/ollama/dist/deps/ ./dist/deps/
+ARG GOFLAGS
+ARG CGO_CFLAGS
+RUN go build -trimpath .
 
-FROM ubuntu:22.04
+# Intermediate stage used for ./scripts/build_linux.sh
+FROM --platform=linux/arm64 cpu-build-arm64 AS build-arm64
+ENV CGO_ENABLED 1
+ARG GOLANG_VERSION
+WORKDIR /go/src/github.com/ollama/ollama
+COPY . .
+COPY --from=static-build-arm64 /go/src/github.com/ollama/ollama/llm/build/linux/ llm/build/linux/
+COPY --from=cuda-build-arm64 /go/src/github.com/ollama/ollama/llm/build/linux/ llm/build/linux/
+ARG GOFLAGS
+ARG CGO_CFLAGS
+RUN go build -trimpath .
+
+# Runtime stages
+FROM --platform=linux/amd64 ubuntu:22.04 as runtime-amd64
 RUN apt-get update && apt-get install -y ca-certificates
-COPY --from=0 /go/src/github.com/jmorganca/ollama/ollama /bin/ollama
+COPY --from=build-amd64 /go/src/github.com/ollama/ollama/ollama /bin/ollama
+FROM --platform=linux/arm64 ubuntu:22.04 as runtime-arm64
+RUN apt-get update && apt-get install -y ca-certificates
+COPY --from=build-arm64 /go/src/github.com/ollama/ollama/ollama /bin/ollama
+
+# Radeon images are much larger so we keep it distinct from the CPU/CUDA image
+FROM --platform=linux/amd64 rocm/dev-centos-7:${ROCM_VERSION}-complete as runtime-rocm
+RUN update-pciids
+COPY --from=build-amd64 /go/src/github.com/ollama/ollama/ollama /bin/ollama
 EXPOSE 11434
 ENV OLLAMA_HOST 0.0.0.0
 
-# set some environment variable for better NVIDIA compatibility
+ENTRYPOINT ["/bin/ollama"]
-ENV PATH=/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
+CMD ["serve"]
+
+FROM runtime-$TARGETARCH
+EXPOSE 11434
+ENV OLLAMA_HOST 0.0.0.0
+ENV PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
 ENV LD_LIBRARY_PATH=/usr/local/nvidia/lib:/usr/local/nvidia/lib64
 ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
+ENV NVIDIA_VISIBLE_DEVICES=all
 
 ENTRYPOINT ["/bin/ollama"]
 CMD ["serve"]

Dockerfile.build

@@ -1,99 +0,0 @@
-ARG GOLANG_VERSION=1.21.3
-ARG CMAKE_VERSION=3.22.1
-ARG CUDA_VERSION=11.3.1
-
-# Copy the minimal context we need to run the generate scripts
-FROM scratch AS llm-code
-COPY .git .git
-COPY .gitmodules .gitmodules
-COPY llm llm
-
-FROM --platform=linux/amd64 nvidia/cuda:$CUDA_VERSION-devel-centos7 AS cuda-build-amd64
-ARG CMAKE_VERSION
-ARG CGO_CFLAGS
-COPY ./scripts/rh_linux_deps.sh /
-RUN CMAKE_VERSION=${CMAKE_VERSION} sh /rh_linux_deps.sh
-ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
-COPY --from=llm-code / /go/src/github.com/jmorganca/ollama/
-WORKDIR /go/src/github.com/jmorganca/ollama/llm/generate
-RUN OLLAMA_SKIP_CPU_GENERATE=1 sh gen_linux.sh
-
-FROM --platform=linux/arm64 nvidia/cuda:$CUDA_VERSION-devel-rockylinux8 AS cuda-build-arm64
-ARG CMAKE_VERSION
-ARG CGO_CFLAGS
-COPY ./scripts/rh_linux_deps.sh /
-RUN CMAKE_VERSION=${CMAKE_VERSION} sh /rh_linux_deps.sh
-ENV PATH /opt/rh/gcc-toolset-10/root/usr/bin:$PATH
-COPY --from=llm-code / /go/src/github.com/jmorganca/ollama/
-WORKDIR /go/src/github.com/jmorganca/ollama/llm/generate
-RUN OLLAMA_SKIP_CPU_GENERATE=1 sh gen_linux.sh
-
-FROM --platform=linux/amd64 rocm/dev-centos-7:5.7.1-complete AS rocm-5-build-amd64
-ARG CMAKE_VERSION
-ARG CGO_CFLAGS
-COPY ./scripts/rh_linux_deps.sh /
-RUN CMAKE_VERSION=${CMAKE_VERSION} sh /rh_linux_deps.sh
-ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
-ENV LIBRARY_PATH /opt/amdgpu/lib64
-COPY --from=llm-code / /go/src/github.com/jmorganca/ollama/
-WORKDIR /go/src/github.com/jmorganca/ollama/llm/generate
-RUN OLLAMA_SKIP_CPU_GENERATE=1 sh gen_linux.sh
-
-FROM --platform=linux/amd64 rocm/dev-centos-7:6.0-complete AS rocm-6-build-amd64
-ARG CMAKE_VERSION
-ARG CGO_CFLAGS
-COPY ./scripts/rh_linux_deps.sh /
-RUN CMAKE_VERSION=${CMAKE_VERSION} sh /rh_linux_deps.sh
-ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
-ENV LIBRARY_PATH /opt/amdgpu/lib64
-COPY --from=llm-code / /go/src/github.com/jmorganca/ollama/
-WORKDIR /go/src/github.com/jmorganca/ollama/llm/generate
-RUN OLLAMA_SKIP_CPU_GENERATE=1 sh gen_linux.sh
-
-FROM --platform=linux/amd64 centos:7 AS cpu-build-amd64
-ARG CMAKE_VERSION
-ARG GOLANG_VERSION
-ARG OLLAMA_CUSTOM_CPU_DEFS
-ARG CGO_CFLAGS
-COPY ./scripts/rh_linux_deps.sh /
-RUN CMAKE_VERSION=${CMAKE_VERSION} GOLANG_VERSION=${GOLANG_VERSION} sh /rh_linux_deps.sh
-ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
-COPY --from=llm-code / /go/src/github.com/jmorganca/ollama/
-WORKDIR /go/src/github.com/jmorganca/ollama/llm/generate
-RUN sh gen_linux.sh
-
-FROM --platform=linux/arm64 centos:7 AS cpu-build-arm64
-ARG CMAKE_VERSION
-ARG GOLANG_VERSION
-ARG OLLAMA_CUSTOM_CPU_DEFS
-ARG CGO_CFLAGS
-COPY ./scripts/rh_linux_deps.sh /
-RUN CMAKE_VERSION=${CMAKE_VERSION} GOLANG_VERSION=${GOLANG_VERSION} sh /rh_linux_deps.sh
-ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
-COPY --from=llm-code / /go/src/github.com/jmorganca/ollama/
-WORKDIR /go/src/github.com/jmorganca/ollama/llm/generate
-RUN sh gen_linux.sh
-
-
-FROM --platform=linux/amd64 cpu-build-amd64 AS build-amd64
-ENV CGO_ENABLED 1
-ARG GOFLAGS
-ARG CGO_CFLAGS
-WORKDIR /go/src/github.com/jmorganca/ollama
-COPY . .
-COPY --from=cuda-build-amd64 /go/src/github.com/jmorganca/ollama/llm/llama.cpp/build/linux/ llm/llama.cpp/build/linux/
-COPY --from=rocm-5-build-amd64 /go/src/github.com/jmorganca/ollama/llm/llama.cpp/build/linux/ llm/llama.cpp/build/linux/
-COPY --from=rocm-6-build-amd64 /go/src/github.com/jmorganca/ollama/llm/llama.cpp/build/linux/ llm/llama.cpp/build/linux/
-RUN go build .
-
-FROM --platform=linux/arm64 cpu-build-arm64 AS build-arm64
-ENV CGO_ENABLED 1
-ARG GOLANG_VERSION
-ARG GOFLAGS
-ARG CGO_CFLAGS
-WORKDIR /go/src/github.com/jmorganca/ollama
-COPY . .
-COPY --from=cuda-build-arm64 /go/src/github.com/jmorganca/ollama/llm/llama.cpp/build/linux/ llm/llama.cpp/build/linux/
-RUN go build .
-
-FROM build-$TARGETARCH

README.md

@@ -1,8 +1,5 @@
 <div align="center">
-  <picture>
+  <img alt="ollama" height="200px" src="https://github.com/ollama/ollama/assets/3325447/0d0b44e2-8f4a-4e99-9b52-a5c1c741c8f7">
-    <source media="(prefers-color-scheme: dark)" height="200px" srcset="https://github.com/jmorganca/ollama/assets/3325447/56ea1849-1284-4645-8970-956de6e51c3c">
-    <img alt="logo" height="200px" src="https://github.com/jmorganca/ollama/assets/3325447/0d0b44e2-8f4a-4e99-9b52-a5c1c741c8f7">
-  </picture>
 </div>
 
 # Ollama
@@ -13,53 +10,57 @@ Get up and running with large language models locally.
 
 ### macOS
 
-[Download](https://ollama.ai/download/Ollama-darwin.zip)
+[Download](https://ollama.com/download/Ollama-darwin.zip)
 
-### Windows
+### Windows preview
 
-Coming soon! For now, you can install Ollama on Windows via WSL2.
+[Download](https://ollama.com/download/OllamaSetup.exe)
 
-### Linux & WSL2
+### Linux
 
 ```
-curl https://ollama.ai/install.sh | sh
+curl -fsSL https://ollama.com/install.sh | sh
 ```
 
-[Manual install instructions](https://github.com/jmorganca/ollama/blob/main/docs/linux.md)
+[Manual install instructions](https://github.com/ollama/ollama/blob/main/docs/linux.md)
 
 ### Docker
 
 The official [Ollama Docker image](https://hub.docker.com/r/ollama/ollama) `ollama/ollama` is available on Docker Hub.
 
+### Libraries
+
+- [ollama-python](https://github.com/ollama/ollama-python)
+- [ollama-js](https://github.com/ollama/ollama-js)
+
 ## Quickstart
 
-To run and chat with [Llama 2](https://ollama.ai/library/llama2):
+To run and chat with [Llama 3](https://ollama.com/library/llama3):
 
 ```
-ollama run llama2
+ollama run llama3
 ```
 
 ## Model library
 
-Ollama supports a list of open-source models available on [ollama.ai/library](https://ollama.ai/library 'ollama model library')
+Ollama supports a list of models available on [ollama.com/library](https://ollama.com/library 'ollama model library')
 
-Here are some example open-source models that can be downloaded:
+Here are some example models that can be downloaded:
 
 | Model              | Parameters | Size  | Download                       |
 | ------------------ | ---------- | ----- | ------------------------------ |
-| Llama 2            | 7B         | 3.8GB | `ollama run llama2`            |
+| Llama 3            | 8B         | 4.7GB | `ollama run llama3`            |
+| Llama 3            | 70B        | 40GB  | `ollama run llama3:70b`        |
+| Phi-3              | 3,8B       | 2.3GB | `ollama run phi3`              |
 | Mistral            | 7B         | 4.1GB | `ollama run mistral`           |
-| Dolphin Phi        | 2.7B       | 1.6GB | `ollama run dolphin-phi`       |
-| Phi-2              | 2.7B       | 1.7GB | `ollama run phi`               |
 | Neural Chat        | 7B         | 4.1GB | `ollama run neural-chat`       |
 | Starling           | 7B         | 4.1GB | `ollama run starling-lm`       |
 | Code Llama         | 7B         | 3.8GB | `ollama run codellama`         |
 | Llama 2 Uncensored | 7B         | 3.8GB | `ollama run llama2-uncensored` |
-| Llama 2 13B        | 13B        | 7.3GB | `ollama run llama2:13b`        |
-| Llama 2 70B        | 70B        | 39GB  | `ollama run llama2:70b`        |
-| Orca Mini          | 3B         | 1.9GB | `ollama run orca-mini`         |
-| Vicuna             | 7B         | 3.8GB | `ollama run vicuna`            |
 | LLaVA              | 7B         | 4.5GB | `ollama run llava`             |
+| Gemma              | 2B         | 1.4GB | `ollama run gemma:2b`          |
+| Gemma              | 7B         | 4.8GB | `ollama run gemma:7b`          |
+| Solar              | 10.7B      | 6.1GB | `ollama run solar`             |
 
 > Note: You should have at least 8 GB of RAM available to run the 7B models, 16 GB to run the 13B models, and 32 GB to run the 33B models.
 
@@ -93,16 +94,16 @@ See the [guide](docs/import.md) on importing models for more information.
 
 ### Customize a prompt
 
-Models from the Ollama library can be customized with a prompt. For example, to customize the `llama2` model:
+Models from the Ollama library can be customized with a prompt. For example, to customize the `llama3` model:
 
 ```
-ollama pull llama2
+ollama pull llama3
 ```
 
 Create a `Modelfile`:
 
 ```
-FROM llama2
+FROM llama3
 
 # set the temperature to 1 [higher is more creative, lower is more coherent]
 PARAMETER temperature 1
@@ -137,7 +138,7 @@ ollama create mymodel -f ./Modelfile
 ### Pull a model
 
 ```
-ollama pull llama2
+ollama pull llama3
 ```
 
 > This command can also be used to update a local model. Only the diff will be pulled.
@@ -145,13 +146,13 @@ ollama pull llama2
 ### Remove a model
 
 ```
-ollama rm llama2
+ollama rm llama3
 ```
 
 ### Copy a model
 
 ```
-ollama cp llama2 my-llama2
+ollama cp llama3 my-model
 ```
 
 ### Multiline input
@@ -175,7 +176,7 @@ The image features a yellow smiley face, which is likely the central focus of th
 ### Pass in prompt as arguments
 
 ```
-$ ollama run llama2 "Summarize this file: $(cat README.md)"
+$ ollama run llama3 "Summarize this file: $(cat README.md)"
  Ollama is a lightweight, extensible framework for building and running language models on the local machine. It provides a simple API for creating, running, and managing models, as well as a library of pre-built models that can be easily used in a variety of applications.
 ```
 
@@ -198,18 +199,21 @@ brew install cmake go
 ```
 
 Then generate dependencies:
+
 ```
 go generate ./...
 ```
+
 Then build the binary:
+
 ```
 go build .
 ```
 
-More detailed instructions can be found in the [developer guide](https://github.com/jmorganca/ollama/blob/main/docs/development.md)
+More detailed instructions can be found in the [developer guide](https://github.com/ollama/ollama/blob/main/docs/development.md)
-
 
 ### Running local builds
+
 Next, start the server:
 
 ```
@@ -219,7 +223,7 @@ Next, start the server:
 Finally, in a separate shell, run a model:
 
 ```
-./ollama run llama2
+./ollama run llama3
 ```
 
 ## REST API
@@ -230,7 +234,7 @@ Ollama has a REST API for running and managing models.
 
 ```
 curl http://localhost:11434/api/generate -d '{
-  "model": "llama2",
+  "model": "llama3",
   "prompt":"Why is the sky blue?"
 }'
 ```
@@ -239,7 +243,7 @@ curl http://localhost:11434/api/generate -d '{
 
 ```
 curl http://localhost:11434/api/chat -d '{
-  "model": "mistral",
+  "model": "llama3",
   "messages": [
     { "role": "user", "content": "why is the sky blue?" }
   ]
@@ -248,26 +252,49 @@ curl http://localhost:11434/api/chat -d '{
 
 See the [API documentation](./docs/api.md) for all endpoints.
 
-## Integrations
-
-- [ollama-python](https://github.com/jmorganca/ollama-python)
-
 ## Community Integrations
 
 ### Web & Desktop
+
+- [Open WebUI](https://github.com/open-webui/open-webui)
+- [Enchanted (macOS native)](https://github.com/AugustDev/enchanted)
+- [Lollms-Webui](https://github.com/ParisNeo/lollms-webui)
+- [LibreChat](https://github.com/danny-avila/LibreChat)
 - [Bionic GPT](https://github.com/bionic-gpt/bionic-gpt)
 - [HTML UI](https://github.com/rtcfirefly/ollama-ui)
+- [Saddle](https://github.com/jikkuatwork/saddle)
 - [Chatbot UI](https://github.com/ivanfioravanti/chatbot-ollama)
+- [Chatbot UI v2](https://github.com/mckaywrigley/chatbot-ui)
 - [Typescript UI](https://github.com/ollama-interface/Ollama-Gui?tab=readme-ov-file)
 - [Minimalistic React UI for Ollama Models](https://github.com/richawo/minimal-llm-ui)
-- [Web UI](https://github.com/ollama-webui/ollama-webui)
 - [Ollamac](https://github.com/kevinhermawan/Ollamac)
-- [big-AGI](https://github.com/enricoros/big-agi/blob/main/docs/config-ollama.md)
+- [big-AGI](https://github.com/enricoros/big-AGI/blob/main/docs/config-local-ollama.md)
 - [Cheshire Cat assistant framework](https://github.com/cheshire-cat-ai/core)
 - [Amica](https://github.com/semperai/amica)
 - [chatd](https://github.com/BruceMacD/chatd)
 - [Ollama-SwiftUI](https://github.com/kghandour/Ollama-SwiftUI)
-
+- [Dify.AI](https://github.com/langgenius/dify)
+- [MindMac](https://mindmac.app)
+- [NextJS Web Interface for Ollama](https://github.com/jakobhoeg/nextjs-ollama-llm-ui)
+- [Msty](https://msty.app)
+- [Chatbox](https://github.com/Bin-Huang/Chatbox)
+- [WinForm Ollama Copilot](https://github.com/tgraupmann/WinForm_Ollama_Copilot)
+- [NextChat](https://github.com/ChatGPTNextWeb/ChatGPT-Next-Web) with [Get Started Doc](https://docs.nextchat.dev/models/ollama)
+- [Alpaca WebUI](https://github.com/mmo80/alpaca-webui)
+- [OllamaGUI](https://github.com/enoch1118/ollamaGUI)
+- [OpenAOE](https://github.com/InternLM/OpenAOE)
+- [Odin Runes](https://github.com/leonid20000/OdinRunes)
+- [LLM-X: Progressive Web App](https://github.com/mrdjohnson/llm-x)
+- [AnythingLLM (Docker + MacOs/Windows/Linux native app)](https://github.com/Mintplex-Labs/anything-llm)
+- [Ollama Basic Chat: Uses HyperDiv Reactive UI](https://github.com/rapidarchitect/ollama_basic_chat)
+- [Ollama-chats RPG](https://github.com/drazdra/ollama-chats)
+- [QA-Pilot: Chat with Code Repository](https://github.com/reid41/QA-Pilot)
+- [ChatOllama: Open Source Chatbot based on Ollama with Knowledge Bases](https://github.com/sugarforever/chat-ollama)
+- [CRAG Ollama Chat: Simple Web Search with Corrective RAG](https://github.com/Nagi-ovo/CRAG-Ollama-Chat)
+- [RAGFlow: Open-source Retrieval-Augmented Generation engine based on deep document understanding](https://github.com/infiniflow/ragflow)
+- [chat: chat web app for teams](https://github.com/swuecho/chat)
+- [Lobe Chat](https://github.com/lobehub/lobe-chat) with [Integrating Doc](https://lobehub.com/docs/self-hosting/examples/ollama)
+- [Ollama RAG Chatbot: Local Chat with multiples PDFs using Ollama and RAG.](https://github.com/datvodinh/rag-chatbot.git)
 
 ### Terminal
 
@@ -276,23 +303,36 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Emacs client](https://github.com/zweifisch/ollama)
 - [gen.nvim](https://github.com/David-Kunz/gen.nvim)
 - [ollama.nvim](https://github.com/nomnivore/ollama.nvim)
+- [ollero.nvim](https://github.com/marco-souza/ollero.nvim)
+- [ollama-chat.nvim](https://github.com/gerazov/ollama-chat.nvim)
 - [ogpt.nvim](https://github.com/huynle/ogpt.nvim)
 - [gptel Emacs client](https://github.com/karthink/gptel)
 - [Oatmeal](https://github.com/dustinblackman/oatmeal)
 - [cmdh](https://github.com/pgibler/cmdh)
+- [ooo](https://github.com/npahlfer/ooo)
+- [shell-pilot](https://github.com/reid41/shell-pilot)
+- [tenere](https://github.com/pythops/tenere)
+- [llm-ollama](https://github.com/taketwo/llm-ollama) for [Datasette's LLM CLI](https://llm.datasette.io/en/stable/).
+- [typechat-cli](https://github.com/anaisbetts/typechat-cli)
+- [ShellOracle](https://github.com/djcopley/ShellOracle)
+- [tlm](https://github.com/yusufcanb/tlm)
+- [podman-ollama](https://github.com/ericcurtin/podman-ollama)
 
 ### Database
 
-- [MindsDB](https://github.com/mindsdb/mindsdb/blob/staging/mindsdb/integrations/handlers/ollama_handler/README.md)
+- [MindsDB](https://github.com/mindsdb/mindsdb/blob/staging/mindsdb/integrations/handlers/ollama_handler/README.md) (Connects Ollama models with nearly 200 data platforms and apps)
+- [chromem-go](https://github.com/philippgille/chromem-go/blob/v0.5.0/embed_ollama.go) with [example](https://github.com/philippgille/chromem-go/tree/v0.5.0/examples/rag-wikipedia-ollama)
 
 ### Package managers
 
 - [Pacman](https://archlinux.org/packages/extra/x86_64/ollama/)
+- [Helm Chart](https://artifacthub.io/packages/helm/ollama-helm/ollama)
 
 ### Libraries
 
 - [LangChain](https://python.langchain.com/docs/integrations/llms/ollama) and [LangChain.js](https://js.langchain.com/docs/modules/model_io/models/llms/integrations/ollama) with [example](https://js.langchain.com/docs/use_cases/question_answering/local_retrieval_qa)
 - [LangChainGo](https://github.com/tmc/langchaingo/) with [example](https://github.com/tmc/langchaingo/tree/main/examples/ollama-completion-example)
+- [LangChain4j](https://github.com/langchain4j/langchain4j) with [example](https://github.com/langchain4j/langchain4j-examples/tree/main/ollama-examples/src/main/java)
 - [LlamaIndex](https://gpt-index.readthedocs.io/en/stable/examples/llm/ollama.html)
 - [LiteLLM](https://github.com/BerriAI/litellm)
 - [OllamaSharp for .NET](https://github.com/awaescher/OllamaSharp)
@@ -306,7 +346,11 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [LangChainDart](https://github.com/davidmigloz/langchain_dart)
 - [Semantic Kernel - Python](https://github.com/microsoft/semantic-kernel/tree/main/python/semantic_kernel/connectors/ai/ollama)
 - [Haystack](https://github.com/deepset-ai/haystack-integrations/blob/main/integrations/ollama.md)
-
+- [Elixir LangChain](https://github.com/brainlid/langchain)
+- [Ollama for R - rollama](https://github.com/JBGruber/rollama)
+- [Ollama-ex for Elixir](https://github.com/lebrunel/ollama-ex)
+- [Ollama Connector for SAP ABAP](https://github.com/b-tocs/abap_btocs_ollama)
+- [Testcontainers](https://testcontainers.com/modules/ollama/)
 
 ### Mobile
 
@@ -320,6 +364,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Continue](https://github.com/continuedev/continue)
 - [Obsidian Ollama plugin](https://github.com/hinterdupfinger/obsidian-ollama)
 - [Logseq Ollama plugin](https://github.com/omagdy7/ollama-logseq)
+- [NotesOllama](https://github.com/andersrex/notesollama) (Apple Notes Ollama plugin)
 - [Dagger Chatbot](https://github.com/samalba/dagger-chatbot)
 - [Discord AI Bot](https://github.com/mekb-turtle/discord-ai-bot)
 - [Ollama Telegram Bot](https://github.com/ruecat/ollama-telegram)
@@ -327,4 +372,16 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Rivet plugin](https://github.com/abrenneke/rivet-plugin-ollama)
 - [Llama Coder](https://github.com/ex3ndr/llama-coder) (Copilot alternative using Ollama)
 - [Obsidian BMO Chatbot plugin](https://github.com/longy2k/obsidian-bmo-chatbot)
-- [Open Interpreter](https://docs.openinterpreter.com/language-model-setup/local-models/ollama)
+- [Cliobot](https://github.com/herval/cliobot) (Telegram bot with Ollama support)
+- [Copilot for Obsidian plugin](https://github.com/logancyang/obsidian-copilot)
+- [Obsidian Local GPT plugin](https://github.com/pfrankov/obsidian-local-gpt)
+- [Open Interpreter](https://docs.openinterpreter.com/language-model-setup/local-models/ollama)
+- [twinny](https://github.com/rjmacarthy/twinny) (Copilot and Copilot chat alternative using Ollama)
+- [Wingman-AI](https://github.com/RussellCanfield/wingman-ai) (Copilot code and chat alternative using Ollama and HuggingFace)
+- [Page Assist](https://github.com/n4ze3m/page-assist) (Chrome Extension)
+- [AI Telegram Bot](https://github.com/tusharhero/aitelegrambot) (Telegram bot using Ollama in backend)
+- [AI ST Completion](https://github.com/yaroslavyaroslav/OpenAI-sublime-text) (Sublime Text 4 AI assistant plugin with Ollama support)
+- [Discord-Ollama Chat Bot](https://github.com/kevinthedang/discord-ollama) (Generalized TypeScript Discord Bot w/ Tuning Documentation)
+
+### Supported backends 
+- [llama.cpp](https://github.com/ggerganov/llama.cpp) project founded by Georgi Gerganov. 

api/client.go

@@ -1,3 +1,9 @@
+// Package api implements the client-side API for code wishing to interact
+// with the ollama service. The methods of the [Client] type correspond to
+// the ollama REST API as described in https://github.com/ollama/ollama/blob/main/docs/api.md
+//
+// The ollama command-line client itself uses this package to interact with
+// the backend service.
 package api
 
 import (
@@ -5,7 +11,6 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
-	"errors"
 	"fmt"
 	"io"
 	"net"
@@ -15,13 +20,15 @@ import (
 	"runtime"
 	"strings"
 
-	"github.com/jmorganca/ollama/format"
+	"github.com/ollama/ollama/format"
-	"github.com/jmorganca/ollama/version"
+	"github.com/ollama/ollama/version"
 )
 
+// Client encapsulates client state for interacting with the ollama
+// service. Use [ClientFromEnvironment] to create new Clients.
 type Client struct {
 	base *url.URL
-	http http.Client
+	http *http.Client
 }
 
 func checkError(resp *http.Response, body []byte) error {
@@ -40,6 +47,15 @@ func checkError(resp *http.Response, body []byte) error {
 	return apiError
 }
 
+// ClientFromEnvironment creates a new [Client] using configuration from the
+// environment variable OLLAMA_HOST, which points to the network host and
+// port on which the ollama service is listenting. The format of this variable
+// is:
+//
+//	<scheme>://<host>:<port>
+//
+// If the variable is not specified, a default ollama host and port will be
+// used.
 func ClientFromEnvironment() (*Client, error) {
 	defaultPort := "11434"
 
@@ -66,30 +82,20 @@ func ClientFromEnvironment() (*Client, error) {
 		}
 	}
 
-	client := Client{
+	return &Client{
 		base: &url.URL{
 			Scheme: scheme,
 			Host:   net.JoinHostPort(host, port),
 		},
-	}
+		http: http.DefaultClient,
+	}, nil
+}
 
-	mockRequest, err := http.NewRequest(http.MethodHead, client.base.String(), nil)
+func NewClient(base *url.URL, http *http.Client) *Client {
-	if err != nil {
+	return &Client{
-		return nil, err
+		base: base,
+		http: http,
 	}
-
-	proxyURL, err := http.ProxyFromEnvironment(mockRequest)
-	if err != nil {
-		return nil, err
-	}
-
-	client.http = http.Client{
-		Transport: &http.Transport{
-			Proxy: http.ProxyURL(proxyURL),
-		},
-	}
-
-	return &client, nil
 }
 
 func (c *Client) do(ctx context.Context, method, path string, reqData, respData any) error {
@@ -208,8 +214,14 @@ func (c *Client) stream(ctx context.Context, method, path string, data any, fn f
 	return nil
 }
 
+// GenerateResponseFunc is a function that [Client.Generate] invokes every time
+// a response is received from the service. If this function returns an error,
+// [Client.Generate] will stop generating and return this error.
 type GenerateResponseFunc func(GenerateResponse) error
 
+// Generate generates a response for a given prompt. The req parameter should
+// be populated with prompt details. fn is called for each response (there may
+// be multiple responses, e.g. in case streaming is enabled).
 func (c *Client) Generate(ctx context.Context, req *GenerateRequest, fn GenerateResponseFunc) error {
 	return c.stream(ctx, http.MethodPost, "/api/generate", req, func(bts []byte) error {
 		var resp GenerateResponse
@@ -221,8 +233,15 @@ func (c *Client) Generate(ctx context.Context, req *GenerateRequest, fn Generate
 	})
 }
 
+// ChatResponseFunc is a function that [Client.Chat] invokes every time
+// a response is received from the service. If this function returns an error,
+// [Client.Chat] will stop generating and return this error.
 type ChatResponseFunc func(ChatResponse) error
 
+// Chat generates the next message in a chat. [ChatRequest] may contain a
+// sequence of messages which can be used to maintain chat history with a model.
+// fn is called for each response (there may be multiple responses, e.g. if case
+// streaming is enabled).
 func (c *Client) Chat(ctx context.Context, req *ChatRequest, fn ChatResponseFunc) error {
 	return c.stream(ctx, http.MethodPost, "/api/chat", req, func(bts []byte) error {
 		var resp ChatResponse
@@ -234,8 +253,14 @@ func (c *Client) Chat(ctx context.Context, req *ChatRequest, fn ChatResponseFunc
 	})
 }
 
+// PullProgressFunc is a function that [Client.Pull] invokes every time there
+// is progress with a "pull" request sent to the service. If this function
+// returns an error, [Client.Pull] will stop the process and return this error.
 type PullProgressFunc func(ProgressResponse) error
 
+// Pull downloads a model from the ollama library. fn is called each time
+// progress is made on the request and can be used to display a progress bar,
+// etc.
 func (c *Client) Pull(ctx context.Context, req *PullRequest, fn PullProgressFunc) error {
 	return c.stream(ctx, http.MethodPost, "/api/pull", req, func(bts []byte) error {
 		var resp ProgressResponse
@@ -318,18 +343,7 @@ func (c *Client) Embeddings(ctx context.Context, req *EmbeddingRequest) (*Embedd
 }
 
 func (c *Client) CreateBlob(ctx context.Context, digest string, r io.Reader) error {
-	if err := c.do(ctx, http.MethodHead, fmt.Sprintf("/api/blobs/%s", digest), nil, nil); err != nil {
+	return c.do(ctx, http.MethodPost, fmt.Sprintf("/api/blobs/%s", digest), r, nil)
-		var statusError StatusError
-		if !errors.As(err, &statusError) || statusError.StatusCode != http.StatusNotFound {
-			return err
-		}
-
-		if err := c.do(ctx, http.MethodPost, fmt.Sprintf("/api/blobs/%s", digest), r, nil); err != nil {
-			return err
-		}
-	}
-
-	return nil
 }
 
 func (c *Client) Version(ctx context.Context) (string, error) {

api/types.go

@@ -2,6 +2,7 @@ package api
 
 import (
 	"encoding/json"
+	"errors"
 	"fmt"
 	"math"
 	"os"
@@ -33,25 +34,55 @@ func (e StatusError) Error() string {
 
 type ImageData []byte
 
+// GenerateRequest describes a request sent by [Client.Generate]. While you
+// have to specify the Model and Prompt fields, all the other fields have
+// reasonable defaults for basic uses.
 type GenerateRequest struct {
-	Model    string      `json:"model"`
+	// Model is the model name; it should be a name familiar to Ollama from
-	Prompt   string      `json:"prompt"`
+	// the library at https://ollama.com/library
-	System   string      `json:"system"`
+	Model string `json:"model"`
-	Template string      `json:"template"`
-	Context  []int       `json:"context,omitempty"`
-	Stream   *bool       `json:"stream,omitempty"`
-	Raw      bool        `json:"raw,omitempty"`
-	Format   string      `json:"format"`
-	Images   []ImageData `json:"images,omitempty"`
 
+	// Prompt is the textual prompt to send to the model.
+	Prompt string `json:"prompt"`
+
+	// System overrides the model's default system message/prompt.
+	System string `json:"system"`
+
+	// Template overrides the model's default prompt template.
+	Template string `json:"template"`
+
+	// Context is the context parameter returned from a previous call to
+	// Generate call. It can be used to keep a short conversational memory.
+	Context []int `json:"context,omitempty"`
+
+	// Stream specifies whether the response is streaming; it is true by default.
+	Stream *bool `json:"stream,omitempty"`
+
+	// Raw set to true means that no formatting will be applied to the prompt.
+	Raw bool `json:"raw,omitempty"`
+
+	// Format specifies the format to return a response in.
+	Format string `json:"format"`
+
+	// KeepAlive controls how long the model will stay loaded in memory following
+	// this request.
+	KeepAlive *Duration `json:"keep_alive,omitempty"`
+
+	// Images is an optional list of base64-encoded images accompanying this
+	// request, for multimodal models.
+	Images []ImageData `json:"images,omitempty"`
+
+	// Options lists model-specific options. For example, temperature can be
+	// set through this field, if the model supports it.
 	Options map[string]interface{} `json:"options"`
 }
 
 type ChatRequest struct {
-	Model    string    `json:"model"`
+	Model     string    `json:"model"`
-	Messages []Message `json:"messages"`
+	Messages  []Message `json:"messages"`
-	Stream   *bool     `json:"stream,omitempty"`
+	Stream    *bool     `json:"stream,omitempty"`
-	Format   string    `json:"format"`
+	Format    string    `json:"format"`
+	KeepAlive *Duration `json:"keep_alive,omitempty"`
 
 	Options map[string]interface{} `json:"options"`
 }
@@ -81,7 +112,7 @@ type Metrics struct {
 	EvalDuration       time.Duration `json:"eval_duration,omitempty"`
 }
 
-// Options specfied in GenerateRequest, if you add a new option here add it to the API docs also
+// Options specified in GenerateRequest, if you add a new option here add it to the API docs also
 type Options struct {
 	Runner
 
@@ -107,27 +138,30 @@ type Options struct {
 
 // Runner options which must be set when the model is loaded into memory
 type Runner struct {
-	UseNUMA            bool    `json:"numa,omitempty"`
+	UseNUMA   bool `json:"numa,omitempty"`
-	NumCtx             int     `json:"num_ctx,omitempty"`
+	NumCtx    int  `json:"num_ctx,omitempty"`
-	NumBatch           int     `json:"num_batch,omitempty"`
+	NumBatch  int  `json:"num_batch,omitempty"`
-	NumGQA             int     `json:"num_gqa,omitempty"`
+	NumGQA    int  `json:"num_gqa,omitempty"`
-	NumGPU             int     `json:"num_gpu,omitempty"`
+	NumGPU    int  `json:"num_gpu,omitempty"`
-	MainGPU            int     `json:"main_gpu,omitempty"`
+	MainGPU   int  `json:"main_gpu,omitempty"`
-	LowVRAM            bool    `json:"low_vram,omitempty"`
+	LowVRAM   bool `json:"low_vram,omitempty"`
-	F16KV              bool    `json:"f16_kv,omitempty"`
+	F16KV     bool `json:"f16_kv,omitempty"`
-	LogitsAll          bool    `json:"logits_all,omitempty"`
+	LogitsAll bool `json:"logits_all,omitempty"`
-	VocabOnly          bool    `json:"vocab_only,omitempty"`
+	VocabOnly bool `json:"vocab_only,omitempty"`
-	UseMMap            bool    `json:"use_mmap,omitempty"`
+	UseMMap   bool `json:"use_mmap,omitempty"`
-	UseMLock           bool    `json:"use_mlock,omitempty"`
+	UseMLock  bool `json:"use_mlock,omitempty"`
-	EmbeddingOnly      bool    `json:"embedding_only,omitempty"`
+	NumThread int  `json:"num_thread,omitempty"`
-	RopeFrequencyBase  float32 `json:"rope_frequency_base,omitempty"`
+
+	// Unused: RopeFrequencyBase is ignored. Instead the value in the model will be used
+	RopeFrequencyBase float32 `json:"rope_frequency_base,omitempty"`
+	// Unused: RopeFrequencyScale is ignored. Instead the value in the model will be used
 	RopeFrequencyScale float32 `json:"rope_frequency_scale,omitempty"`
-	NumThread          int     `json:"num_thread,omitempty"`
 }
 
 type EmbeddingRequest struct {
-	Model  string `json:"model"`
+	Model     string    `json:"model"`
-	Prompt string `json:"prompt"`
+	Prompt    string    `json:"prompt"`
+	KeepAlive *Duration `json:"keep_alive,omitempty"`
 
 	Options map[string]interface{} `json:"options"`
 }
@@ -137,10 +171,11 @@ type EmbeddingResponse struct {
 }
 
 type CreateRequest struct {
-	Model     string `json:"model"`
+	Model        string `json:"model"`
-	Path      string `json:"path"`
+	Path         string `json:"path"`
-	Modelfile string `json:"modelfile"`
+	Modelfile    string `json:"modelfile"`
-	Stream    *bool  `json:"stream,omitempty"`
+	Stream       *bool  `json:"stream,omitempty"`
+	Quantization string `json:"quantization,omitempty"`
 
 	// Name is deprecated, see Model
 	Name string `json:"name"`
@@ -171,6 +206,7 @@ type ShowResponse struct {
 	Template   string       `json:"template,omitempty"`
 	System     string       `json:"system,omitempty"`
 	Details    ModelDetails `json:"details,omitempty"`
+	Messages   []Message    `json:"messages,omitempty"`
 }
 
 type CopyRequest struct {
@@ -236,6 +272,7 @@ type GenerateResponse struct {
 }
 
 type ModelDetails struct {
+	ParentModel       string   `json:"parent_model"`
 	Format            string   `json:"format"`
 	Family            string   `json:"family"`
 	Families          []string `json:"families"`
@@ -271,7 +308,7 @@ func (m *Metrics) Summary() {
 	}
 }
 
-var ErrInvalidOpts = fmt.Errorf("invalid options")
+var ErrInvalidOpts = errors.New("invalid options")
 
 func (opts *Options) FromMap(m map[string]interface{}) error {
 	valueOpts := reflect.ValueOf(opts).Elem() // names of the fields in the options struct
@@ -359,8 +396,10 @@ func (opts *Options) FromMap(m map[string]interface{}) error {
 func DefaultOptions() Options {
 	return Options{
 		// options set on request to runner
-		NumPredict:       -1,
+		NumPredict: -1,
-		NumKeep:          0,
+
+		// set a minimal num_keep to avoid issues on context shifts
+		NumKeep:          4,
 		Temperature:      0.8,
 		TopK:             40,
 		TopP:             0.9,
@@ -378,19 +417,16 @@ func DefaultOptions() Options {
 
 		Runner: Runner{
 			// options set when the model is loaded
-			NumCtx:             2048,
+			NumCtx:    2048,
-			RopeFrequencyBase:  10000.0,
+			NumBatch:  512,
-			RopeFrequencyScale: 1.0,
+			NumGPU:    -1, // -1 here indicates that NumGPU should be set dynamically
-			NumBatch:           512,
+			NumGQA:    1,
-			NumGPU:             -1, // -1 here indicates that NumGPU should be set dynamically
+			NumThread: 0, // let the runtime decide
-			NumGQA:             1,
+			LowVRAM:   false,
-			NumThread:          0, // let the runtime decide
+			F16KV:     true,
-			LowVRAM:            false,
+			UseMLock:  false,
-			F16KV:              true,
+			UseMMap:   true,
-			UseMLock:           false,
+			UseNUMA:   false,
-			UseMMap:            true,
-			UseNUMA:            false,
-			EmbeddingOnly:      true,
 		},
 	}
 }
@@ -410,15 +446,18 @@ func (d *Duration) UnmarshalJSON(b []byte) (err error) {
 	switch t := v.(type) {
 	case float64:
 		if t < 0 {
-			t = math.MaxFloat64
+			d.Duration = time.Duration(math.MaxInt64)
+		} else {
+			d.Duration = time.Duration(t * float64(time.Second))
 		}
-
-		d.Duration = time.Duration(t)
 	case string:
 		d.Duration, err = time.ParseDuration(t)
 		if err != nil {
 			return err
 		}
+		if d.Duration < 0 {
+			d.Duration = time.Duration(math.MaxInt64)
+		}
 	}
 
 	return nil

api/types_test.go

@@ -0,0 +1,50 @@
+package api
+
+import (
+	"encoding/json"
+	"math"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestKeepAliveParsingFromJSON(t *testing.T) {
+	tests := []struct {
+		name string
+		req  string
+		exp  *Duration
+	}{
+		{
+			name: "Positive Integer",
+			req:  `{ "keep_alive": 42 }`,
+			exp:  &Duration{42 * time.Second},
+		},
+		{
+			name: "Positive Integer String",
+			req:  `{ "keep_alive": "42m" }`,
+			exp:  &Duration{42 * time.Minute},
+		},
+		{
+			name: "Negative Integer",
+			req:  `{ "keep_alive": -1 }`,
+			exp:  &Duration{math.MaxInt64},
+		},
+		{
+			name: "Negative Integer String",
+			req:  `{ "keep_alive": "-1m" }`,
+			exp:  &Duration{math.MaxInt64},
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			var dec ChatRequest
+			err := json.Unmarshal([]byte(test.req), &dec)
+			require.NoError(t, err)
+
+			assert.Equal(t, test.exp, dec.KeepAlive)
+		})
+	}
+}

app/.eslintrc.json

@@ -1,16 +0,0 @@
-{
-  "env": {
-    "browser": true,
-    "es6": true,
-    "node": true
-  },
-  "extends": [
-    "eslint:recommended",
-    "plugin:@typescript-eslint/eslint-recommended",
-    "plugin:@typescript-eslint/recommended",
-    "plugin:import/recommended",
-    "plugin:import/electron",
-    "plugin:import/typescript"
-  ],
-  "parser": "@typescript-eslint/parser"
-}

app/.gitignore

@@ -1,92 +1,2 @@
-# Logs
+ollama.syso
-logs
+app
-*.log
-npm-debug.log*
-yarn-debug.log*
-yarn-error.log*
-lerna-debug.log*
-
-# Diagnostic reports (https://nodejs.org/api/report.html)
-report.[0-9]*.[0-9]*.[0-9]*.[0-9]*.json
-
-# Runtime data
-pids
-*.pid
-*.seed
-*.pid.lock
-.DS_Store
-
-# Directory for instrumented libs generated by jscoverage/JSCover
-lib-cov
-
-# Coverage directory used by tools like istanbul
-coverage
-*.lcov
-
-# nyc test coverage
-.nyc_output
-
-# node-waf configuration
-.lock-wscript
-
-# Compiled binary addons (https://nodejs.org/api/addons.html)
-build/Release
-
-# Dependency directories
-node_modules/
-jspm_packages/
-
-# TypeScript v1 declaration files
-typings/
-
-# TypeScript cache
-*.tsbuildinfo
-
-# Optional npm cache directory
-.npm
-
-# Optional eslint cache
-.eslintcache
-
-# Optional REPL history
-.node_repl_history
-
-# Output of 'npm pack'
-*.tgz
-
-# Yarn Integrity file
-.yarn-integrity
-
-# dotenv environment variables file
-.env
-.env.test
-
-# parcel-bundler cache (https://parceljs.org/)
-.cache
-
-# next.js build output
-.next
-
-# nuxt.js build output
-.nuxt
-
-# vuepress build output
-.vuepress/dist
-
-# Serverless directories
-.serverless/
-
-# FuseBox cache
-.fusebox/
-
-# DynamoDB Local files
-.dynamodb/
-
-# Webpack
-.webpack/
-
-# Vite
-.vite/
-
-# Electron-Forge
-out/

app/AppDelegate.h

@@ -0,0 +1,7 @@
+#import <Cocoa/Cocoa.h>
+
+@interface AppDelegate : NSObject <NSApplicationDelegate>
+
+- (void)applicationDidFinishLaunching:(NSNotification *)aNotification;
+
+@end

app/README.md

@@ -1,21 +1,18 @@
-# Desktop
+# Ollama App
 
-This app builds upon Ollama to provide a desktop experience for running models.
+## macOS
 
-## Developing
+TODO
 
-First, build the `ollama` binary:
+## Windows
+
+If you want to build the installer, youll need to install
+- https://jrsoftware.org/isinfo.php
+
+
+In the top directory of this repo, run the following powershell script
+to build the ollama CLI, ollama app, and ollama installer.
 
 ```
-cd ..
+powershell -ExecutionPolicy Bypass -File .\scripts\build_windows.ps1
-go build .
 ```
-
-Then run the desktop app with `npm start`:
-
-```
-cd app
-npm install
-npm start
-```
-

app/app_darwin.go

@@ -0,0 +1,76 @@
+package main
+
+// #cgo CFLAGS: -x objective-c
+// #cgo LDFLAGS: -framework Cocoa -framework LocalAuthentication -framework ServiceManagement
+// #include "app_darwin.h"
+import "C"
+import (
+	"context"
+	"fmt"
+	"log/slog"
+	"os"
+	"path/filepath"
+	"syscall"
+)
+
+func init() {
+	home, err := os.UserHomeDir()
+	if err != nil {
+		panic(err)
+	}
+
+	ServerLogFile = filepath.Join(home, ".ollama", "logs", "server.log")
+}
+
+func run() {
+	initLogging()
+	slog.Info("ollama macOS app started")
+
+	// Ask to move to applications directory
+	moving := C.askToMoveToApplications()
+	if moving {
+		return
+	}
+
+	C.killOtherInstances()
+
+	code := C.installSymlink()
+	if code != 0 {
+		slog.Error("Failed to install symlink")
+	}
+
+	exe, err := os.Executable()
+	if err != nil {
+		panic(err)
+	}
+
+	var options ServerOptions
+
+	ctx, cancel := context.WithCancel(context.Background())
+	var done chan int
+
+	done, err = SpawnServer(ctx, filepath.Join(filepath.Dir(exe), "..", "Resources", "ollama"), options)
+	if err != nil {
+		slog.Error(fmt.Sprintf("Failed to spawn ollama server %s", err))
+		done = make(chan int, 1)
+		done <- 1
+	}
+
+	// Run the native macOS app
+	// Note: this will block until the app is closed
+	C.run()
+
+	slog.Info("ollama macOS app closed")
+
+	cancel()
+	slog.Info("Waiting for ollama server to shutdown...")
+	if done != nil {
+		<-done
+	}
+	slog.Info("Ollama app exiting")
+}
+
+//export Quit
+func Quit() {
+	syscall.Kill(os.Getpid(), syscall.SIGTERM)
+}

app/app_darwin.h

@@ -0,0 +1,13 @@
+#import <Cocoa/Cocoa.h>
+
+@interface AppDelegate : NSObject <NSApplicationDelegate>
+- (void)applicationDidFinishLaunching:(NSNotification *)aNotification;
+@end
+
+void run();
+void killOtherInstances();
+bool askToMoveToApplications();
+int createSymlinkWithAuthorization();
+int installSymlink();
+extern void Restart();
+extern void Quit();

app/app_darwin.m

@@ -0,0 +1,282 @@
+#import <AppKit/AppKit.h>
+#import <Cocoa/Cocoa.h>
+#import <CoreServices/CoreServices.h>
+#import <Security/Security.h>
+#import <ServiceManagement/ServiceManagement.h>
+#import "app_darwin.h"
+
+@interface AppDelegate ()
+
+@property (strong, nonatomic) NSStatusItem *statusItem;
+
+@end
+
+@implementation AppDelegate
+
+- (void)applicationDidFinishLaunching:(NSNotification *)aNotification {
+    // show status menu
+    NSMenu *menu = [[NSMenu alloc] init];
+
+    NSMenuItem *aboutMenuItem = [[NSMenuItem alloc] initWithTitle:@"About Ollama" action:@selector(aboutOllama) keyEquivalent:@""];
+    [aboutMenuItem setTarget:self];
+    [menu addItem:aboutMenuItem];
+
+    // Settings submenu
+    NSMenu *settingsMenu = [[NSMenu alloc] initWithTitle:@"Settings"];
+
+    // Submenu items
+    NSMenuItem *chooseModelDirectoryItem = [[NSMenuItem alloc] initWithTitle:@"Choose model directory..." action:@selector(chooseModelDirectory) keyEquivalent:@""];
+    [chooseModelDirectoryItem setTarget:self];
+    [chooseModelDirectoryItem setEnabled:YES];
+    [settingsMenu addItem:chooseModelDirectoryItem];
+
+    NSMenuItem *exposeExternallyItem = [[NSMenuItem alloc] initWithTitle:@"Allow external connections" action:@selector(toggleExposeExternally:) keyEquivalent:@""];
+    [exposeExternallyItem setTarget:self];
+    [exposeExternallyItem setState:NSOffState]; // Set initial state to off
+    [exposeExternallyItem setEnabled:YES];
+    [settingsMenu addItem:exposeExternallyItem];
+
+    NSMenuItem *allowCrossOriginItem = [[NSMenuItem alloc] initWithTitle:@"Allow browser requests" action:@selector(toggleCrossOrigin:) keyEquivalent:@""];
+    [allowCrossOriginItem setTarget:self];
+    [allowCrossOriginItem setState:NSOffState]; // Set initial state to off
+    [allowCrossOriginItem setEnabled:YES];
+    [settingsMenu addItem:allowCrossOriginItem];
+
+    NSMenuItem *settingsMenuItem = [[NSMenuItem alloc] initWithTitle:@"Settings" action:nil keyEquivalent:@""];
+    [settingsMenuItem setSubmenu:settingsMenu];
+    [menu addItem:settingsMenuItem];
+
+    [menu addItemWithTitle:@"Quit Ollama" action:@selector(quit) keyEquivalent:@"q"];
+
+    self.statusItem = [[NSStatusBar systemStatusBar] statusItemWithLength:NSVariableStatusItemLength];
+    [self.statusItem addObserver:self forKeyPath:@"button.effectiveAppearance" options:NSKeyValueObservingOptionNew|NSKeyValueObservingOptionInitial context:nil];
+
+    self.statusItem.menu = menu;
+    [self showIcon];
+}
+
+- (void)aboutOllama {
+    [[NSApplication sharedApplication] orderFrontStandardAboutPanel:nil];
+}
+
+- (void)toggleCrossOrigin:(id)sender {
+    NSMenuItem *item = (NSMenuItem *)sender;
+    if ([item state] == NSOffState) {
+        // Do something when cross-origin requests are allowed
+        [item setState:NSOnState];
+    } else {
+        // Do something when cross-origin requests are disallowed
+        [item setState:NSOffState];
+    }
+}
+
+- (void)toggleExposeExternally:(id)sender {
+    NSMenuItem *item = (NSMenuItem *)sender;
+    if ([item state] == NSOffState) {
+        // Do something when Ollama is exposed externally
+        [item setState:NSOnState];
+    } else {
+        // Do something when Ollama is not exposed externally
+        [item setState:NSOffState];
+    }
+}
+
+- (void)chooseModelDirectory {
+    NSOpenPanel *openPanel = [NSOpenPanel openPanel];
+    [openPanel setCanChooseFiles:NO];
+    [openPanel setCanChooseDirectories:YES];
+    [openPanel setAllowsMultipleSelection:NO];
+
+    NSInteger result = [openPanel runModal];
+    if (result == NSModalResponseOK) {
+        NSURL *selectedDirectoryURL = [openPanel URLs].firstObject;
+        // Do something with the selected directory URL
+    }
+}
+
+-(void) showIcon {
+    NSAppearance* appearance = self.statusItem.button.effectiveAppearance;
+    NSString* appearanceName = (NSString*)(appearance.name);
+    NSString* iconName = [[appearanceName lowercaseString] containsString:@"dark"] ? @"iconDark" : @"icon";
+    NSImage* statusImage = [NSImage imageNamed:iconName];
+    [statusImage setTemplate:YES];
+    self.statusItem.button.image = statusImage;
+}
+
+-(void)observeValueForKeyPath:(NSString *)keyPath ofObject:(id)object change:(NSDictionary<NSKeyValueChangeKey,id> *)change context:(void *)context {
+    [self showIcon];
+}
+
+- (void)quit {
+    [NSApp stop:nil];
+}
+
+@end
+
+void run() {
+    @autoreleasepool {
+        [NSApplication sharedApplication];
+        AppDelegate *appDelegate = [[AppDelegate alloc] init];
+        [NSApp setDelegate:appDelegate];
+        [NSApp run];
+    }
+}
+
+// killOtherInstances kills all other instances of the app currently
+// running. This way we can ensure that only the most recently started
+// instance of Ollama is running
+void killOtherInstances() {
+    pid_t pid = getpid();
+    NSArray *all = [[NSWorkspace sharedWorkspace] runningApplications];
+    NSMutableArray *apps = [NSMutableArray array];
+
+    for (NSRunningApplication *app in all) {
+        if ([app.bundleIdentifier isEqualToString:[[NSBundle mainBundle] bundleIdentifier]] ||
+            [app.bundleIdentifier isEqualToString:@"ai.ollama.ollama"] ||
+            [app.bundleIdentifier isEqualToString:@"com.electron.ollama"]) {
+            if (app.processIdentifier != pid) {
+                [apps addObject:app];
+            }
+        }
+    }
+
+    for (NSRunningApplication *app in apps) {
+        kill(app.processIdentifier, SIGTERM);
+    }
+
+    NSDate *startTime = [NSDate date];
+    for (NSRunningApplication *app in apps) {
+        while (!app.terminated) {
+            if (-[startTime timeIntervalSinceNow] >= 5) {
+                kill(app.processIdentifier, SIGKILL);
+                break;
+            }
+
+            [[NSRunLoop currentRunLoop] runUntilDate:[NSDate dateWithTimeIntervalSinceNow:0.1]];
+        }
+    }
+}
+
+bool askToMoveToApplications() {
+    NSString *bundlePath = [[NSBundle mainBundle] bundlePath];
+    if ([bundlePath hasPrefix:@"/Applications"]) {
+        return false;
+    }
+
+    NSAlert *alert = [[NSAlert alloc] init];
+    [alert setMessageText:@"Move to Applications?"];
+    [alert setInformativeText:@"Ollama works best when run from the Applications directory."];
+    [alert addButtonWithTitle:@"Move to Applications"];
+    [alert addButtonWithTitle:@"Don't move"];
+
+    [NSApp activateIgnoringOtherApps:YES];
+
+    if ([alert runModal] != NSAlertFirstButtonReturn) {
+        return false;
+    }
+
+    // move to applications
+    NSString *applicationsPath = @"/Applications";
+    NSString *newPath = [applicationsPath stringByAppendingPathComponent:@"Ollama.app"];
+    NSFileManager *fileManager = [NSFileManager defaultManager];
+
+    // Check if the newPath already exists
+    if ([fileManager fileExistsAtPath:newPath]) {
+        NSError *removeError = nil;
+        [fileManager removeItemAtPath:newPath error:&removeError];
+        if (removeError) {
+            NSLog(@"Error removing file at %@: %@", newPath, removeError);
+            return false; // or handle the error
+        }
+    }
+
+    NSError *moveError = nil;
+    [fileManager moveItemAtPath:bundlePath toPath:newPath error:&moveError];
+    if (moveError) {
+        NSLog(@"Error moving file from %@ to %@: %@", bundlePath, newPath, moveError);
+        return false;
+    }
+
+    NSLog(@"Opening %@", newPath);
+    NSError *error = nil;
+    NSWorkspace *workspace = [NSWorkspace sharedWorkspace];
+#pragma clang diagnostic ignored "-Wdeprecated-declarations"
+    [workspace launchApplicationAtURL:[NSURL fileURLWithPath:newPath]
+               options:NSWorkspaceLaunchNewInstance | NSWorkspaceLaunchDefault
+               configuration:@{}
+               error:&error];
+
+    return true;
+}
+
+int installSymlink() {
+    NSString *linkPath = @"/usr/local/bin/ollama";
+    NSError *error = nil;
+
+    NSFileManager *fileManager = [NSFileManager defaultManager];
+    NSString *symlinkPath = [fileManager destinationOfSymbolicLinkAtPath:linkPath error:&error];
+    NSString *bundlePath = [[NSBundle mainBundle] bundlePath];
+    NSString *execPath = [[NSBundle mainBundle] executablePath];
+    NSString *resPath = [[NSBundle mainBundle] pathForResource:@"ollama" ofType:nil];
+
+    // if the symlink already exists and points to the right place, don't prompt
+    if ([symlinkPath isEqualToString:resPath]) {
+        NSLog(@"symbolic link already exists and points to the right place");
+        return 0;
+    }
+
+    NSString *authorizationPrompt = @"Ollama is trying to install its command line interface (CLI) tool.";
+
+    AuthorizationRef auth = NULL;
+    OSStatus createStatus = AuthorizationCreate(NULL, kAuthorizationEmptyEnvironment, kAuthorizationFlagDefaults, &auth);
+    if (createStatus != errAuthorizationSuccess) {
+        NSLog(@"Error creating authorization");
+        return -1;
+    }
+
+    NSString * bundleIdentifier = [[NSBundle mainBundle] bundleIdentifier];
+    NSString *rightNameString = [NSString stringWithFormat:@"%@.%@", bundleIdentifier, @"auth3"];
+    const char *rightName = rightNameString.UTF8String;
+
+    OSStatus getRightResult = AuthorizationRightGet(rightName, NULL);
+    if (getRightResult == errAuthorizationDenied) {
+        if (AuthorizationRightSet(auth, rightName, (__bridge CFTypeRef _Nonnull)(@(kAuthorizationRuleAuthenticateAsAdmin)), (__bridge CFStringRef _Nullable)(authorizationPrompt), NULL, NULL) != errAuthorizationSuccess) {
+            NSLog(@"Failed to set right");
+            return -1;
+        }
+    }
+
+    AuthorizationItem right = { .name = rightName, .valueLength = 0, .value = NULL, .flags = 0 };
+    AuthorizationRights rights = { .count = 1, .items = &right };
+    AuthorizationFlags flags = (AuthorizationFlags)(kAuthorizationFlagExtendRights | kAuthorizationFlagInteractionAllowed);
+    AuthorizationItem iconAuthorizationItem = {.name = kAuthorizationEnvironmentIcon, .valueLength = 0, .value = NULL, .flags = 0};
+    AuthorizationEnvironment authorizationEnvironment = {.count = 0, .items = NULL};
+
+    BOOL failedToUseSystemDomain = NO;
+    OSStatus copyStatus = AuthorizationCopyRights(auth, &rights, &authorizationEnvironment, flags, NULL);
+    if (copyStatus != errAuthorizationSuccess) {
+        failedToUseSystemDomain = YES;
+
+        if (copyStatus == errAuthorizationCanceled) {
+            NSLog(@"User cancelled authorization");
+            return -1;
+        } else {
+            NSLog(@"Failed copying system domain rights: %d", copyStatus);
+            return -1;
+        }
+    }
+
+    const char *toolPath = "/bin/ln";
+    const char *args[] = {"-s", "-F", [resPath UTF8String], "/usr/local/bin/ollama", NULL};
+    FILE *pipe = NULL;
+
+#pragma clang diagnostic ignored "-Wdeprecated-declarations"
+    OSStatus status = AuthorizationExecuteWithPrivileges(auth, toolPath, kAuthorizationFlagDefaults, (char *const *)args, &pipe);
+    if (status != errAuthorizationSuccess) {
+        NSLog(@"Failed to create symlink");
+        return -1;
+    }
+
+    AuthorizationFree(auth, kAuthorizationFlagDestroyRights);
+    return 0;
+}

app/app_windows.go

@@ -0,0 +1,166 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"log"
+	"log/slog"
+	"os"
+	"os/exec"
+	"os/signal"
+	"path/filepath"
+	"strings"
+	"syscall"
+
+	"github.com/ollama/ollama/app/lifecycle"
+	"github.com/ollama/ollama/app/store"
+	"github.com/ollama/ollama/app/tray"
+	"github.com/ollama/ollama/app/updater"
+)
+
+func init() {
+	AppName += ".exe"
+	CLIName += ".exe"
+	// Logs, configs, downloads go to LOCALAPPDATA
+	localAppData := os.Getenv("LOCALAPPDATA")
+	AppDataDir = filepath.Join(localAppData, "Ollama")
+	AppLogFile = filepath.Join(AppDataDir, "app.log")
+	ServerLogFile = filepath.Join(AppDataDir, "server.log")
+
+	// Executables are stored in APPDATA
+	AppDir = filepath.Join(localAppData, "Programs", "Ollama")
+
+	// Make sure we have PATH set correctly for any spawned children
+	paths := strings.Split(os.Getenv("PATH"), ";")
+	// Start with whatever we find in the PATH/LD_LIBRARY_PATH
+	found := false
+	for _, path := range paths {
+		d, err := filepath.Abs(path)
+		if err != nil {
+			continue
+		}
+		if strings.EqualFold(AppDir, d) {
+			found = true
+		}
+	}
+	if !found {
+		paths = append(paths, AppDir)
+
+		pathVal := strings.Join(paths, ";")
+		slog.Debug("setting PATH=" + pathVal)
+		err := os.Setenv("PATH", pathVal)
+		if err != nil {
+			slog.Error(fmt.Sprintf("failed to update PATH: %s", err))
+		}
+	}
+
+	// Make sure our logging dir exists
+	_, err := os.Stat(AppDataDir)
+	if errors.Is(err, os.ErrNotExist) {
+		if err := os.MkdirAll(AppDataDir, 0o755); err != nil {
+			slog.Error(fmt.Sprintf("create ollama dir %s: %v", AppDataDir, err))
+		}
+	}
+}
+
+func ShowLogs() {
+	cmd_path := "c:\\Windows\\system32\\cmd.exe"
+	slog.Debug(fmt.Sprintf("viewing logs with start %s", AppDataDir))
+	cmd := exec.Command(cmd_path, "/c", "start", AppDataDir)
+	cmd.SysProcAttr = &syscall.SysProcAttr{HideWindow: false, CreationFlags: 0x08000000}
+	err := cmd.Start()
+	if err != nil {
+		slog.Error(fmt.Sprintf("Failed to open log dir: %s", err))
+	}
+}
+
+func Start() {
+	cmd_path := "c:\\Windows\\system32\\cmd.exe"
+	slog.Debug(fmt.Sprintf("viewing logs with start %s", AppDataDir))
+	cmd := exec.Command(cmd_path, "/c", "start", AppDataDir)
+	cmd.SysProcAttr = &syscall.SysProcAttr{HideWindow: false, CreationFlags: 0x08000000}
+	err := cmd.Start()
+	if err != nil {
+		slog.Error(fmt.Sprintf("Failed to open log dir: %s", err))
+	}
+}
+
+func run() {
+	initLogging()
+
+	slog.Info("ollama windows app started")
+
+	ctx, cancel := context.WithCancel(context.Background())
+	var done chan int
+
+	t, err := tray.NewTray()
+	if err != nil {
+		log.Fatalf("Failed to start: %s", err)
+	}
+	callbacks := t.GetCallbacks()
+
+	signals := make(chan os.Signal, 1)
+	signal.Notify(signals, syscall.SIGINT, syscall.SIGTERM)
+
+	go func() {
+		slog.Debug("starting callback loop")
+		for {
+			select {
+			case <-callbacks.Quit:
+				slog.Debug("quit called")
+				t.Quit()
+			case <-signals:
+				slog.Debug("shutting down due to signal")
+				t.Quit()
+			case <-callbacks.Update:
+				err := updater.DoUpgrade(cancel, done)
+				if err != nil {
+					slog.Warn(fmt.Sprintf("upgrade attempt failed: %s", err))
+				}
+			case <-callbacks.ShowLogs:
+				ShowLogs()
+			case <-callbacks.DoFirstUse:
+				err := lifecycle.GetStarted()
+				if err != nil {
+					slog.Warn(fmt.Sprintf("Failed to launch getting started shell: %s", err))
+				}
+			}
+		}
+	}()
+
+	if !store.GetFirstTimeRun() {
+		slog.Debug("First time run")
+		err = t.DisplayFirstUseNotification()
+		if err != nil {
+			slog.Debug(fmt.Sprintf("XXX failed to display first use notification %v", err))
+		}
+		store.SetFirstTimeRun(true)
+	} else {
+		slog.Debug("Not first time, skipping first run notification")
+	}
+
+	if isServerRunning(ctx) {
+		slog.Info("Detected another instance of ollama running, exiting")
+		os.Exit(1)
+	}
+
+	done, err = SpawnServer(ctx, CLIName)
+	if err != nil {
+		// TODO - should we retry in a backoff loop?
+		// TODO - should we pop up a warning and maybe add a menu item to view application logs?
+		slog.Error(fmt.Sprintf("Failed to spawn ollama server %s", err))
+		done = make(chan int, 1)
+		done <- 1
+	}
+
+	updater.StartBackgroundUpdaterChecker(ctx, t.UpdateAvailable)
+
+	t.Run()
+	cancel()
+	slog.Info("Waiting for ollama server to shutdown...")
+	if done != nil {
+		<-done
+	}
+	slog.Info("Ollama app exiting")
+}

app/assets/assets.go

@@ -0,0 +1,17 @@
+package assets
+
+import (
+	"embed"
+	"io/fs"
+)
+
+//go:embed *.ico
+var icons embed.FS
+
+func ListIcons() ([]string, error) {
+	return fs.Glob(icons, "*")
+}
+
+func GetIcon(filename string) ([]byte, error) {
+	return icons.ReadFile(filename)
+}

app/darwin/Ollama.app/Contents/Info.plist

@@ -0,0 +1,40 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+  <dict>
+    <key>CFBundleDisplayName</key>
+    <string>Ollama</string>
+    <key>CFBundleExecutable</key>
+    <string>Ollama</string>
+    <key>CFBundleIconFile</key>
+    <string>icon.icns</string>
+    <key>CFBundleIdentifier</key>
+    <string>com.ollama.ollama</string>
+    <key>CFBundleInfoDictionaryVersion</key>
+    <string>6.0</string>
+    <key>CFBundleName</key>
+    <string>Ollama</string>
+    <key>CFBundlePackageType</key>
+    <string>APPL</string>
+    <key>CFBundleShortVersionString</key>
+    <string>0.0.0</string>
+    <key>CFBundleVersion</key>
+    <string>0.0.0</string>
+    <key>DTCompiler</key>
+    <string>com.apple.compilers.llvm.clang.1_0</string>
+    <key>DTSDKBuild</key>
+    <string>22E245</string>
+    <key>DTSDKName</key>
+    <string>macosx13.3</string>
+    <key>DTXcode</key>
+    <string>1431</string>
+    <key>DTXcodeBuild</key>
+    <string>14E300c</string>
+    <key>LSApplicationCategoryType</key>
+    <string>public.app-category.developer-tools</string>
+    <key>LSMinimumSystemVersion</key>
+    <string>11.0</string>
+    <key>LSUIElement</key>
+    <true/>
+  </dict>
+</plist>

app/forge.config.ts

@@ -1,78 +0,0 @@
-import type { ForgeConfig } from '@electron-forge/shared-types'
-import { MakerSquirrel } from '@electron-forge/maker-squirrel'
-import { MakerZIP } from '@electron-forge/maker-zip'
-import { PublisherGithub } from '@electron-forge/publisher-github'
-import { AutoUnpackNativesPlugin } from '@electron-forge/plugin-auto-unpack-natives'
-import { WebpackPlugin } from '@electron-forge/plugin-webpack'
-import * as path from 'path'
-import * as fs from 'fs'
-
-import { mainConfig } from './webpack.main.config'
-import { rendererConfig } from './webpack.renderer.config'
-
-const packageJson = JSON.parse(fs.readFileSync(path.resolve(__dirname, './package.json'), 'utf8'))
-
-const config: ForgeConfig = {
-  packagerConfig: {
-    appVersion: process.env.VERSION || packageJson.version,
-    asar: true,
-    icon: './assets/icon.icns',
-    extraResource: [
-      '../dist/ollama',
-      path.join(__dirname, './assets/iconTemplate.png'),
-      path.join(__dirname, './assets/iconTemplate@2x.png'),
-      path.join(__dirname, './assets/iconUpdateTemplate.png'),
-      path.join(__dirname, './assets/iconUpdateTemplate@2x.png'),
-      path.join(__dirname, './assets/iconDarkTemplate.png'),
-      path.join(__dirname, './assets/iconDarkTemplate@2x.png'),
-      path.join(__dirname, './assets/iconDarkUpdateTemplate.png'),
-      path.join(__dirname, './assets/iconDarkUpdateTemplate@2x.png'),
-    ],
-    ...(process.env.SIGN
-      ? {
-          osxSign: {
-            identity: process.env.APPLE_IDENTITY,
-          },
-          osxNotarize: {
-            tool: 'notarytool',
-            appleId: process.env.APPLE_ID || '',
-            appleIdPassword: process.env.APPLE_PASSWORD || '',
-            teamId: process.env.APPLE_TEAM_ID || '',
-          },
-        }
-      : {}),
-    osxUniversal: {
-      x64ArchFiles: '**/ollama',
-    },
-  },
-  rebuildConfig: {},
-  makers: [new MakerSquirrel({}), new MakerZIP({}, ['darwin'])],
-  hooks: {
-    readPackageJson: async (_, packageJson) => {
-      return { ...packageJson, version: process.env.VERSION || packageJson.version }
-    },
-  },
-  plugins: [
-    new AutoUnpackNativesPlugin({}),
-    new WebpackPlugin({
-      mainConfig,
-      devContentSecurityPolicy: `default-src * 'unsafe-eval' 'unsafe-inline'; img-src data: 'self'`,
-      renderer: {
-        config: rendererConfig,
-        nodeIntegration: true,
-        entryPoints: [
-          {
-            html: './src/index.html',
-            js: './src/renderer.tsx',
-            name: 'main_window',
-            preload: {
-              js: './src/preload.ts',
-            },
-          },
-        ],
-      },
-    }),
-  ],
-}
-
-export default config

app/lifecycle/getstarted_darwin.go

@@ -0,0 +1,7 @@
+package lifecycle
+
+import "fmt"
+
+func GetStarted() error {
+	return fmt.Errorf("GetStarted not implemented")
+}

app/lifecycle/getstarted_windows.go

@@ -0,0 +1,44 @@
+package lifecycle
+
+import (
+	"fmt"
+	"log/slog"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"syscall"
+)
+
+func GetStarted() error {
+	const CREATE_NEW_CONSOLE = 0x00000010
+	var err error
+	bannerScript := filepath.Join(AppDir, "ollama_welcome.ps1")
+	args := []string{
+		// TODO once we're signed, the execution policy bypass should be removed
+		"powershell", "-noexit", "-ExecutionPolicy", "Bypass", "-nologo", "-file", bannerScript,
+	}
+	args[0], err = exec.LookPath(args[0])
+	if err != nil {
+		return err
+	}
+
+	// Make sure the script actually exists
+	_, err = os.Stat(bannerScript)
+	if err != nil {
+		return fmt.Errorf("getting started banner script error %s", err)
+	}
+
+	slog.Info(fmt.Sprintf("opening getting started terminal with %v", args))
+	attrs := &os.ProcAttr{
+		Files: []*os.File{os.Stdin, os.Stdout, os.Stderr},
+		Sys:   &syscall.SysProcAttr{CreationFlags: CREATE_NEW_CONSOLE, HideWindow: false},
+	}
+	proc, err := os.StartProcess(args[0], args, attrs)
+
+	if err != nil {
+		return fmt.Errorf("unable to start getting started shell %w", err)
+	}
+
+	slog.Debug(fmt.Sprintf("getting started terminal PID: %d", proc.Pid))
+	return proc.Release()
+}

app/lifecycle/paths.go

@@ -0,0 +1,74 @@
+package lifecycle
+
+import (
+	"errors"
+	"fmt"
+	"log/slog"
+	"os"
+	"path/filepath"
+	"runtime"
+	"strings"
+)
+
+var (
+	AppName    = "ollama app"
+	CLIName    = "ollama"
+	AppDir     = "/opt/Ollama"
+	AppDataDir = "/opt/Ollama"
+	// TODO - should there be a distinct log dir?
+	UpdateStageDir = "/tmp"
+	AppLogFile     = "/tmp/ollama_app.log"
+	ServerLogFile  = "/tmp/ollama.log"
+	UpgradeLogFile = "/tmp/ollama_update.log"
+	Installer      = "OllamaSetup.exe"
+)
+
+func init() {
+	if runtime.GOOS == "windows" {
+		AppName += ".exe"
+		CLIName += ".exe"
+		// Logs, configs, downloads go to LOCALAPPDATA
+		localAppData := os.Getenv("LOCALAPPDATA")
+		AppDataDir = filepath.Join(localAppData, "Ollama")
+		UpdateStageDir = filepath.Join(AppDataDir, "updates")
+		AppLogFile = filepath.Join(AppDataDir, "app.log")
+		ServerLogFile = filepath.Join(AppDataDir, "server.log")
+		UpgradeLogFile = filepath.Join(AppDataDir, "upgrade.log")
+
+		// Executables are stored in APPDATA
+		AppDir = filepath.Join(localAppData, "Programs", "Ollama")
+
+		// Make sure we have PATH set correctly for any spawned children
+		paths := strings.Split(os.Getenv("PATH"), ";")
+		// Start with whatever we find in the PATH/LD_LIBRARY_PATH
+		found := false
+		for _, path := range paths {
+			d, err := filepath.Abs(path)
+			if err != nil {
+				continue
+			}
+			if strings.EqualFold(AppDir, d) {
+				found = true
+			}
+		}
+		if !found {
+			paths = append(paths, AppDir)
+
+			pathVal := strings.Join(paths, ";")
+			slog.Debug("setting PATH=" + pathVal)
+			err := os.Setenv("PATH", pathVal)
+			if err != nil {
+				slog.Error(fmt.Sprintf("failed to update PATH: %s", err))
+			}
+		}
+
+		// Make sure our logging dir exists
+		_, err := os.Stat(AppDataDir)
+		if errors.Is(err, os.ErrNotExist) {
+			if err := os.MkdirAll(AppDataDir, 0o755); err != nil {
+				slog.Error(fmt.Sprintf("create ollama dir %s: %v", AppDataDir, err))
+			}
+		}
+
+	}
+}

app/log.go

@@ -0,0 +1,44 @@
+package main
+
+import (
+	"fmt"
+	"log/slog"
+	"os"
+	"path/filepath"
+)
+
+func initLogging() {
+	level := slog.LevelInfo
+
+	if debug := os.Getenv("OLLAMA_DEBUG"); debug != "" {
+		level = slog.LevelDebug
+	}
+
+	var logFile *os.File
+	var err error
+	// Detect if we're a GUI app on windows, and if not, send logs to console
+	if os.Stderr.Fd() != 0 {
+		// Console app detected
+		logFile = os.Stderr
+		// TODO - write one-line to the app.log file saying we're running in console mode to help avoid confusion
+	} else {
+		logFile, err = os.OpenFile(AppLogFile, os.O_APPEND|os.O_WRONLY|os.O_CREATE, 0755)
+		if err != nil {
+			slog.Error(fmt.Sprintf("failed to create server log %v", err))
+			return
+		}
+	}
+	handler := slog.NewTextHandler(logFile, &slog.HandlerOptions{
+		Level:     level,
+		AddSource: true,
+		ReplaceAttr: func(_ []string, attr slog.Attr) slog.Attr {
+			if attr.Key == slog.SourceKey {
+				source := attr.Value.Any().(*slog.Source)
+				source.File = filepath.Base(source.File)
+			}
+			return attr
+		},
+	})
+
+	slog.SetDefault(slog.New(handler))
+}

app/main.go

@@ -0,0 +1,16 @@
+package main
+
+// Compile with the following to get rid of the cmd pop up on windows
+// go build -ldflags="-H windowsgui" .
+var (
+	AppName       string
+	CLIName       string
+	AppDir        string
+	AppDataDir    string
+	AppLogFile    string
+	ServerLogFile string
+)
+
+func main() {
+	run()
+}

app/package.json

@@ -1,84 +0,0 @@
-{
-  "name": "ollama",
-  "productName": "Ollama",
-  "version": "0.0.0",
-  "description": "ollama",
-  "main": ".webpack/main",
-  "scripts": {
-    "start": "electron-forge start",
-    "package": "electron-forge package --arch universal",
-    "package:sign": "SIGN=1 electron-forge package --arch universal",
-    "make": "electron-forge make --arch universal",
-    "make:sign": "SIGN=1 electron-forge make --arch universal",
-    "publish": "SIGN=1 electron-forge publish",
-    "lint": "eslint --ext .ts,.tsx .",
-    "format": "prettier --check . --ignore-path .gitignore",
-    "format:fix": "prettier --write . --ignore-path .gitignore"
-  },
-  "keywords": [],
-  "author": {
-    "name": "Jeffrey Morgan",
-    "email": "jmorganca@gmail.com"
-  },
-  "license": "MIT",
-  "devDependencies": {
-    "@babel/core": "^7.22.5",
-    "@babel/preset-react": "^7.22.5",
-    "@electron-forge/cli": "^6.2.1",
-    "@electron-forge/maker-deb": "^6.2.1",
-    "@electron-forge/maker-rpm": "^6.2.1",
-    "@electron-forge/maker-squirrel": "^6.2.1",
-    "@electron-forge/maker-zip": "^6.2.1",
-    "@electron-forge/plugin-auto-unpack-natives": "^6.2.1",
-    "@electron-forge/plugin-webpack": "^6.2.1",
-    "@electron-forge/publisher-github": "^6.2.1",
-    "@electron/universal": "^1.4.1",
-    "@svgr/webpack": "^8.0.1",
-    "@types/chmodr": "^1.0.0",
-    "@types/node": "^20.4.0",
-    "@types/react": "^18.2.14",
-    "@types/react-dom": "^18.2.6",
-    "@types/uuid": "^9.0.2",
-    "@typescript-eslint/eslint-plugin": "^5.60.0",
-    "@typescript-eslint/parser": "^5.60.0",
-    "@vercel/webpack-asset-relocator-loader": "^1.7.3",
-    "babel-loader": "^9.1.2",
-    "chmodr": "^1.2.0",
-    "copy-webpack-plugin": "^11.0.0",
-    "css-loader": "^6.8.1",
-    "electron": "25.9.2",
-    "eslint": "^8.43.0",
-    "eslint-plugin-import": "^2.27.5",
-    "fork-ts-checker-webpack-plugin": "^7.3.0",
-    "node-loader": "^2.0.0",
-    "postcss": "^8.4.24",
-    "postcss-import": "^15.1.0",
-    "postcss-loader": "^7.3.3",
-    "postcss-preset-env": "^8.5.1",
-    "prettier": "^2.8.8",
-    "prettier-plugin-tailwindcss": "^0.3.0",
-    "style-loader": "^3.3.3",
-    "svg-inline-loader": "^0.8.2",
-    "tailwindcss": "^3.3.2",
-    "ts-loader": "^9.4.3",
-    "ts-node": "^10.9.1",
-    "typescript": "~4.5.4",
-    "url-loader": "^4.1.1",
-    "webpack": "^5.88.0",
-    "webpack-cli": "^5.1.4",
-    "webpack-dev-server": "^4.15.1"
-  },
-  "dependencies": {
-    "@electron/remote": "^2.0.10",
-    "@heroicons/react": "^2.0.18",
-    "@segment/analytics-node": "^1.0.0",
-    "copy-to-clipboard": "^3.3.3",
-    "electron-squirrel-startup": "^1.0.0",
-    "electron-store": "^8.1.0",
-    "react": "^18.2.0",
-    "react-dom": "^18.2.0",
-    "uuid": "^9.0.0",
-    "winston": "^3.10.0",
-    "winston-daily-rotate-file": "^4.7.1"
-  }
-}

app/postcss.config.js

@@ -1,7 +0,0 @@
-module.exports = {
-  plugins: {
-    'postcss-import': {},
-    tailwindcss: {},
-    autoprefixer: {},
-  },
-}

app/server.go

@@ -0,0 +1,163 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"log/slog"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"time"
+
+	"github.com/ollama/ollama/api"
+)
+
+type ServerOptions struct {
+	Cors       bool
+	Expose     bool
+	ModelsPath string
+}
+
+func start(ctx context.Context, command string, options ServerOptions) (*exec.Cmd, error) {
+	cmd := getCmd(ctx, command)
+
+	// set environment variables
+	if options.ModelsPath != "" {
+		cmd.Env = append(cmd.Env, fmt.Sprintf("OLLAMA_MODELS=%s", options.ModelsPath))
+	}
+
+	if options.Cors {
+		cmd.Env = append(cmd.Env, "OLLAMA_ORIGINS=*")
+	}
+
+	if options.Expose {
+		cmd.Env = append(cmd.Env, "OLLAMA_HOST=0.0.0.0")
+	}
+
+	stdout, err := cmd.StdoutPipe()
+	if err != nil {
+		return nil, fmt.Errorf("failed to spawn server stdout pipe: %w", err)
+	}
+	stderr, err := cmd.StderrPipe()
+	if err != nil {
+		return nil, fmt.Errorf("failed to spawn server stderr pipe: %w", err)
+	}
+
+	// TODO - rotation
+	logFile, err := os.OpenFile(ServerLogFile, os.O_APPEND|os.O_WRONLY|os.O_CREATE, 0755)
+	if err != nil {
+		return nil, fmt.Errorf("failed to create server log: %w", err)
+	}
+	go func() {
+		defer logFile.Close()
+		io.Copy(logFile, stdout) //nolint:errcheck
+	}()
+	go func() {
+		defer logFile.Close()
+		io.Copy(logFile, stderr) //nolint:errcheck
+	}()
+
+	// Re-wire context done behavior to attempt a graceful shutdown of the server
+	cmd.Cancel = func() error {
+		if cmd.Process != nil {
+			err := terminate(cmd)
+			if err != nil {
+				slog.Warn("error trying to gracefully terminate server", "err", err)
+				return cmd.Process.Kill()
+			}
+
+			tick := time.NewTicker(10 * time.Millisecond)
+			defer tick.Stop()
+
+			for {
+				select {
+				case <-tick.C:
+					exited, err := isProcessExited(cmd.Process.Pid)
+					if err != nil {
+						return err
+					}
+
+					if exited {
+						return nil
+					}
+				case <-time.After(5 * time.Second):
+					slog.Warn("graceful server shutdown timeout, killing", "pid", cmd.Process.Pid)
+					return cmd.Process.Kill()
+				}
+			}
+		}
+		return nil
+	}
+
+	// run the command and wait for it to finish
+	if err := cmd.Start(); err != nil {
+		return nil, fmt.Errorf("failed to start server %w", err)
+	}
+	if cmd.Process != nil {
+		slog.Info(fmt.Sprintf("started ollama server with pid %d", cmd.Process.Pid))
+	}
+	slog.Info(fmt.Sprintf("ollama server logs %s", ServerLogFile))
+
+	return cmd, nil
+}
+
+func SpawnServer(ctx context.Context, command string, options ServerOptions) (chan int, error) {
+	logDir := filepath.Dir(ServerLogFile)
+	_, err := os.Stat(logDir)
+	if errors.Is(err, os.ErrNotExist) {
+		if err := os.MkdirAll(logDir, 0o755); err != nil {
+			return nil, fmt.Errorf("create ollama server log dir %s: %v", logDir, err)
+		}
+	}
+
+	done := make(chan int)
+
+	go func() {
+		// Keep the server running unless we're shuttind down the app
+		crashCount := 0
+		for {
+			slog.Info(fmt.Sprintf("starting server..."))
+			cmd, err := start(ctx, command, options)
+			if err != nil {
+				slog.Error(fmt.Sprintf("failed to start server %s", err))
+			}
+
+			cmd.Wait() //nolint:errcheck
+			var code int
+			if cmd.ProcessState != nil {
+				code = cmd.ProcessState.ExitCode()
+			}
+
+			select {
+			case <-ctx.Done():
+				slog.Info(fmt.Sprintf("server shutdown with exit code %d", code))
+				done <- code
+				return
+			default:
+				crashCount++
+				slog.Warn(fmt.Sprintf("server crash %d - exit code %d - respawning", crashCount, code))
+				time.Sleep(500 * time.Millisecond * time.Duration(crashCount))
+				break
+			}
+		}
+	}()
+
+	return done, nil
+}
+
+func isServerRunning(ctx context.Context) bool {
+	client, err := api.ClientFromEnvironment()
+	if err != nil {
+		slog.Info("unable to connect to server")
+		return false
+	}
+	err = client.Heartbeat(ctx)
+	if err != nil {
+		slog.Debug(fmt.Sprintf("heartbeat from server: %s", err))
+		slog.Info("unable to connect to server")
+		return false
+	}
+	return true
+}

app/server_darwin.go

@@ -0,0 +1,36 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"os/exec"
+	"syscall"
+)
+
+func getCmd(ctx context.Context, cmd string) *exec.Cmd {
+	return exec.CommandContext(ctx, cmd, "serve")
+}
+
+func terminate(cmd *exec.Cmd) error {
+	return cmd.Process.Signal(os.Interrupt)
+}
+
+func isProcessExited(pid int) (bool, error) {
+	proc, err := os.FindProcess(pid)
+	if err != nil {
+		return false, fmt.Errorf("failed to find process: %v", err)
+	}
+
+	err = proc.Signal(syscall.Signal(0))
+	if err != nil {
+		if errors.Is(err, os.ErrProcessDone) || errors.Is(err, syscall.ESRCH) {
+			return true, nil
+		}
+
+		return false, fmt.Errorf("error signaling process: %v", err)
+	}
+
+	return false, nil
+}

app/server_windows.go

@@ -0,0 +1,89 @@
+package main
+
+import (
+	"context"
+	"fmt"
+	"os/exec"
+	"syscall"
+
+	"golang.org/x/sys/windows"
+)
+
+func getCmd(ctx context.Context, exePath string) *exec.Cmd {
+	cmd := exec.CommandContext(ctx, exePath, "serve")
+	cmd.SysProcAttr = &syscall.SysProcAttr{
+		HideWindow:    true,
+		CreationFlags: windows.CREATE_NEW_PROCESS_GROUP,
+	}
+
+	return cmd
+}
+
+func terminate(cmd *exec.Cmd) error {
+	dll, err := windows.LoadDLL("kernel32.dll")
+	if err != nil {
+		return err
+	}
+	defer dll.Release() // nolint: errcheck
+
+	pid := cmd.Process.Pid
+
+	f, err := dll.FindProc("AttachConsole")
+	if err != nil {
+		return err
+	}
+
+	r1, _, err := f.Call(uintptr(pid))
+	if r1 == 0 && err != syscall.ERROR_ACCESS_DENIED {
+		return err
+	}
+
+	f, err = dll.FindProc("SetConsoleCtrlHandler")
+	if err != nil {
+		return err
+	}
+
+	r1, _, err = f.Call(0, 1)
+	if r1 == 0 {
+		return err
+	}
+
+	f, err = dll.FindProc("GenerateConsoleCtrlEvent")
+	if err != nil {
+		return err
+	}
+
+	r1, _, err = f.Call(windows.CTRL_BREAK_EVENT, uintptr(pid))
+	if r1 == 0 {
+		return err
+	}
+
+	r1, _, err = f.Call(windows.CTRL_C_EVENT, uintptr(pid))
+	if r1 == 0 {
+		return err
+	}
+
+	return nil
+}
+
+const STILL_ACTIVE = 259
+
+func isProcessExited(pid int) (bool, error) {
+	hProcess, err := windows.OpenProcess(windows.PROCESS_QUERY_INFORMATION, false, uint32(pid))
+	if err != nil {
+		return false, fmt.Errorf("failed to open process: %v", err)
+	}
+	defer windows.CloseHandle(hProcess) // nolint: errcheck
+
+	var exitCode uint32
+	err = windows.GetExitCodeProcess(hProcess, &exitCode)
+	if err != nil {
+		return false, fmt.Errorf("failed to get exit code: %v", err)
+	}
+
+	if exitCode == STILL_ACTIVE {
+		return false, nil
+	}
+
+	return true, nil
+}

app/src/app.css

@@ -1,34 +0,0 @@
-@tailwind base;
-@tailwind components;
-@tailwind utilities;
-
-html,
-body {
-  background: transparent;
-}
-
-.drag {
-  -webkit-app-region: drag;
-}
-
-.no-drag {
-  -webkit-app-region: no-drag;
-}
-
-.blink {
-  -webkit-animation: 1s blink step-end infinite;
-  -moz-animation: 1s blink step-end infinite;
-  -ms-animation: 1s blink step-end infinite;
-  -o-animation: 1s blink step-end infinite;
-  animation: 1s blink step-end infinite;
-}
-
-@keyframes blink {
-  from,
-  to {
-    color: transparent;
-  }
-  50% {
-    color: black;
-  }
-}

app/src/app.tsx

@@ -1,122 +0,0 @@
-import { useState } from 'react'
-import copy from 'copy-to-clipboard'
-import { CheckIcon, DocumentDuplicateIcon } from '@heroicons/react/24/outline'
-import Store from 'electron-store'
-import { getCurrentWindow, app } from '@electron/remote'
-
-import { install } from './install'
-import OllamaIcon from './ollama.svg'
-
-const store = new Store()
-
-enum Step {
-  WELCOME = 0,
-  CLI,
-  FINISH,
-}
-
-export default function () {
-  const [step, setStep] = useState<Step>(Step.WELCOME)
-  const [commandCopied, setCommandCopied] = useState<boolean>(false)
-
-  const command = 'ollama run llama2'
-
-  return (
-    <div className='drag'>
-      <div className='mx-auto flex min-h-screen w-full flex-col justify-between bg-white px-4 pt-16'>
-        {step === Step.WELCOME && (
-          <>
-            <div className='mx-auto text-center'>
-              <h1 className='mb-6 mt-4 text-2xl tracking-tight text-gray-900'>Welcome to Ollama</h1>
-              <p className='mx-auto w-[65%] text-sm text-gray-400'>
-                Let's get you up and running with your own large language models.
-              </p>
-              <button
-                onClick={() => setStep(Step.CLI)}
-                className='no-drag rounded-dm mx-auto my-8 w-[40%] rounded-md bg-black px-4 py-2 text-sm text-white hover:brightness-110'
-              >
-                Next
-              </button>
-            </div>
-            <div className='mx-auto'>
-              <OllamaIcon />
-            </div>
-          </>
-        )}
-        {step === Step.CLI && (
-          <>
-            <div className='mx-auto flex flex-col space-y-28 text-center'>
-              <h1 className='mt-4 text-2xl tracking-tight text-gray-900'>Install the command line</h1>
-              <pre className='mx-auto text-4xl text-gray-400'>&gt; ollama</pre>
-              <div className='mx-auto'>
-                <button
-                  onClick={async () => {
-                    try {
-                      await install()
-                      setStep(Step.FINISH)
-                    } catch (e) {
-                      console.error('could not install: ', e)
-                    } finally {
-                      getCurrentWindow().show()
-                      getCurrentWindow().focus()
-                    }
-                  }}
-                  className='no-drag rounded-dm mx-auto w-[60%] rounded-md bg-black px-4 py-2 text-sm text-white hover:brightness-110'
-                >
-                  Install
-                </button>
-                <p className='mx-auto my-4 w-[70%] text-xs text-gray-400'>
-                  You will be prompted for administrator access
-                </p>
-              </div>
-            </div>
-          </>
-        )}
-        {step === Step.FINISH && (
-          <>
-            <div className='mx-auto flex flex-col space-y-20 text-center'>
-              <h1 className='mt-4 text-2xl tracking-tight text-gray-900'>Run your first model</h1>
-              <div className='flex flex-col'>
-                <div className='group relative flex items-center'>
-                  <pre className='language-none text-2xs w-full rounded-md bg-gray-100 px-4 py-3 text-start leading-normal'>
-                    {command}
-                  </pre>
-                  <button
-                    className={`no-drag absolute right-[5px] px-2 py-2 ${
-                      commandCopied
-                        ? 'text-gray-900 opacity-100 hover:cursor-auto'
-                        : 'text-gray-200 opacity-50 hover:cursor-pointer'
-                    } hover:font-bold hover:text-gray-900 group-hover:opacity-100`}
-                    onClick={() => {
-                      copy(command)
-                      setCommandCopied(true)
-                      setTimeout(() => setCommandCopied(false), 3000)
-                    }}
-                  >
-                    {commandCopied ? (
-                      <CheckIcon className='h-4 w-4 font-bold text-gray-500' />
-                    ) : (
-                      <DocumentDuplicateIcon className='h-4 w-4 text-gray-500' />
-                    )}
-                  </button>
-                </div>
-                <p className='mx-auto my-4 w-[70%] text-xs text-gray-400'>
-                  Run this command in your favorite terminal.
-                </p>
-              </div>
-              <button
-                onClick={() => {
-                  store.set('first-time-run', true)
-                  window.close()
-                }}
-                className='no-drag rounded-dm mx-auto w-[60%] rounded-md bg-black px-4 py-2 text-sm text-white hover:brightness-110'
-              >
-                Finish
-              </button>
-            </div>
-          </>
-        )}
-      </div>
-    </div>
-  )
-}

app/src/declarations.d.ts

@@ -1,4 +0,0 @@
-declare module '*.svg' {
-  const content: string
-  export default content
-}

app/src/index.html

@@ -1,9 +0,0 @@
-<!DOCTYPE html>
-<html>
-  <head>
-    <meta charset="UTF-8" />
-  </head>
-  <body>
-    <div id="app"></div>
-  </body>
-</html>

app/src/index.ts

@@ -1,302 +0,0 @@
-import { spawn, ChildProcess } from 'child_process'
-import { app, autoUpdater, dialog, Tray, Menu, BrowserWindow, MenuItemConstructorOptions, nativeTheme } from 'electron'
-import Store from 'electron-store'
-import winston from 'winston'
-import 'winston-daily-rotate-file'
-import * as path from 'path'
-
-import { v4 as uuidv4 } from 'uuid'
-import { installed } from './install'
-
-require('@electron/remote/main').initialize()
-
-if (require('electron-squirrel-startup')) {
-  app.quit()
-}
-
-const store = new Store()
-
-let welcomeWindow: BrowserWindow | null = null
-
-declare const MAIN_WINDOW_WEBPACK_ENTRY: string
-
-const logger = winston.createLogger({
-  transports: [
-    new winston.transports.Console(),
-    new winston.transports.File({
-      filename: path.join(app.getPath('home'), '.ollama', 'logs', 'server.log'),
-      maxsize: 1024 * 1024 * 20,
-      maxFiles: 5,
-    }),
-  ],
-  format: winston.format.printf(info => info.message),
-})
-
-app.on('ready', () => {
-  const gotTheLock = app.requestSingleInstanceLock()
-  if (!gotTheLock) {
-    app.exit(0)
-    return
-  }
-
-  app.on('second-instance', () => {
-    if (app.hasSingleInstanceLock()) {
-      app.releaseSingleInstanceLock()
-    }
-
-    if (proc) {
-      proc.off('exit', restart)
-      proc.kill()
-    }
-
-    app.exit(0)
-  })
-
-  app.focus({ steal: true })
-
-  init()
-})
-
-function firstRunWindow() {
-  // Create the browser window.
-  welcomeWindow = new BrowserWindow({
-    width: 400,
-    height: 500,
-    frame: false,
-    fullscreenable: false,
-    resizable: false,
-    movable: true,
-    show: false,
-    webPreferences: {
-      nodeIntegration: true,
-      contextIsolation: false,
-    },
-  })
-
-  require('@electron/remote/main').enable(welcomeWindow.webContents)
-
-  welcomeWindow.loadURL(MAIN_WINDOW_WEBPACK_ENTRY)
-  welcomeWindow.on('ready-to-show', () => welcomeWindow.show())
-  welcomeWindow.on('closed', () => {
-    if (process.platform === 'darwin') {
-      app.dock.hide()
-    }
-  })
-}
-
-let tray: Tray | null = null
-let updateAvailable = false
-const assetPath = app.isPackaged ? process.resourcesPath : path.join(__dirname, '..', '..', 'assets')
-
-function trayIconPath() {
-  return nativeTheme.shouldUseDarkColors
-    ? updateAvailable
-      ? path.join(assetPath, 'iconDarkUpdateTemplate.png')
-      : path.join(assetPath, 'iconDarkTemplate.png')
-    : updateAvailable
-    ? path.join(assetPath, 'iconUpdateTemplate.png')
-    : path.join(assetPath, 'iconTemplate.png')
-}
-
-function updateTrayIcon() {
-  if (tray) {
-    tray.setImage(trayIconPath())
-  }
-}
-
-function updateTray() {
-  const updateItems: MenuItemConstructorOptions[] = [
-    { label: 'An update is available', enabled: false },
-    {
-      label: 'Restart to update',
-      click: () => autoUpdater.quitAndInstall(),
-    },
-    { type: 'separator' },
-  ]
-
-  const menu = Menu.buildFromTemplate([
-    ...(updateAvailable ? updateItems : []),
-    { role: 'quit', label: 'Quit Ollama', accelerator: 'Command+Q' },
-  ])
-
-  if (!tray) {
-    tray = new Tray(trayIconPath())
-  }
-
-  tray.setToolTip(updateAvailable ? 'An update is available' : 'Ollama')
-  tray.setContextMenu(menu)
-  tray.setImage(trayIconPath())
-
-  nativeTheme.off('updated', updateTrayIcon)
-  nativeTheme.on('updated', updateTrayIcon)
-}
-
-let proc: ChildProcess = null
-
-function server() {
-  const binary = app.isPackaged
-    ? path.join(process.resourcesPath, 'ollama')
-    : path.resolve(process.cwd(), '..', 'ollama')
-
-  proc = spawn(binary, ['serve'])
-
-  proc.stdout.on('data', data => {
-    logger.info(data.toString().trim())
-  })
-
-  proc.stderr.on('data', data => {
-    logger.error(data.toString().trim())
-  })
-
-  proc.on('exit', restart)
-}
-
-function restart() {
-  setTimeout(server, 1000)
-}
-
-app.on('before-quit', () => {
-  if (proc) {
-    proc.off('exit', restart)
-    proc.kill('SIGINT') // send SIGINT signal to the server, which also stops any loaded llms
-  }
-})
-
-const updateURL = `https://ollama.ai/api/update?os=${process.platform}&arch=${
-  process.arch
-}&version=${app.getVersion()}&id=${id()}`
-
-let latest = ''
-async function isNewReleaseAvailable() {
-  try {
-    const response = await fetch(updateURL)
-
-    if (!response.ok) {
-      return false
-    }
-
-    if (response.status === 204) {
-      return false
-    }
-
-    const data = await response.json()
-
-    const url = data?.url
-    if (!url) {
-      return false
-    }
-
-    if (latest === url) {
-      return false
-    }
-
-    latest = url
-
-    return true
-  } catch (error) {
-    logger.error(`update check failed - ${error}`)
-    return false
-  }
-}
-
-async function checkUpdate() {
-  const available = await isNewReleaseAvailable()
-  if (available) {
-    logger.info('checking for update')
-    autoUpdater.checkForUpdates()
-  }
-}
-
-function init() {
-  if (app.isPackaged) {
-    checkUpdate()
-    setInterval(() => {
-      checkUpdate()
-    }, 60 * 60 * 1000)
-  }
-
-  updateTray()
-
-  if (process.platform === 'darwin') {
-    if (app.isPackaged) {
-      if (!app.isInApplicationsFolder()) {
-        const chosen = dialog.showMessageBoxSync({
-          type: 'question',
-          buttons: ['Move to Applications', 'Do Not Move'],
-          message: 'Ollama works best when run from the Applications directory.',
-          defaultId: 0,
-          cancelId: 1,
-        })
-
-        if (chosen === 0) {
-          try {
-            app.moveToApplicationsFolder({
-              conflictHandler: conflictType => {
-                if (conflictType === 'existsAndRunning') {
-                  dialog.showMessageBoxSync({
-                    type: 'info',
-                    message: 'Cannot move to Applications directory',
-                    detail:
-                      'Another version of Ollama is currently running from your Applications directory. Close it first and try again.',
-                  })
-                }
-                return true
-              },
-            })
-            return
-          } catch (e) {
-            logger.error(`[Move to Applications] Failed to move to applications folder - ${e.message}}`)
-          }
-        }
-      }
-    }
-  }
-
-  server()
-
-  if (store.get('first-time-run') && installed()) {
-    if (process.platform === 'darwin') {
-      app.dock.hide()
-    }
-
-    app.setLoginItemSettings({ openAtLogin: app.getLoginItemSettings().openAtLogin })
-    return
-  }
-
-  // This is the first run or the CLI is no longer installed
-  app.setLoginItemSettings({ openAtLogin: true })
-  firstRunWindow()
-}
-
-// Quit when all windows are closed, except on macOS. There, it's common
-// for applications and their menu bar to stay active until the user quits
-// explicitly with Cmd + Q.
-app.on('window-all-closed', () => {
-  if (process.platform !== 'darwin') {
-    app.quit()
-  }
-})
-
-function id(): string {
-  const id = store.get('id') as string
-
-  if (id) {
-    return id
-  }
-
-  const uuid = uuidv4()
-  store.set('id', uuid)
-  return uuid
-}
-
-autoUpdater.setFeedURL({ url: updateURL })
-
-autoUpdater.on('error', e => {
-  logger.error(`update check failed - ${e.message}`)
-  console.error(`update check failed - ${e.message}`)
-})
-
-autoUpdater.on('update-downloaded', () => {
-  updateAvailable = true
-  updateTray()
-})

app/src/install.ts

@@ -1,21 +0,0 @@
-import * as fs from 'fs'
-import { exec as cbExec } from 'child_process'
-import * as path from 'path'
-import { promisify } from 'util'
-
-const app = process && process.type === 'renderer' ? require('@electron/remote').app : require('electron').app
-const ollama = app.isPackaged ? path.join(process.resourcesPath, 'ollama') : path.resolve(process.cwd(), '..', 'ollama')
-const exec = promisify(cbExec)
-const symlinkPath = '/usr/local/bin/ollama'
-
-export function installed() {
-  return fs.existsSync(symlinkPath) && fs.readlinkSync(symlinkPath) === ollama
-}
-
-export async function install() {
-  const command = `do shell script "mkdir -p ${path.dirname(
-    symlinkPath
-  )} && ln -F -s \\"${ollama}\\" \\"${symlinkPath}\\"" with administrator privileges`
-
-  await exec(`osascript -e '${command}'`)
-}

app/src/renderer.tsx

@@ -1,7 +0,0 @@
-import App from './app'
-import './app.css'
-import { createRoot } from 'react-dom/client'
-
-const container = document.getElementById('app')
-const root = createRoot(container)
-root.render(<App />)

app/store/store.go

@@ -0,0 +1,98 @@
+package store
+
+import (
+	"encoding/json"
+	"errors"
+	"fmt"
+	"log/slog"
+	"os"
+	"path/filepath"
+	"sync"
+
+	"github.com/google/uuid"
+)
+
+type Store struct {
+	ID           string `json:"id"`
+	FirstTimeRun bool   `json:"first-time-run"`
+}
+
+var (
+	lock  sync.Mutex
+	store Store
+)
+
+func GetID() string {
+	lock.Lock()
+	defer lock.Unlock()
+	if store.ID == "" {
+		initStore()
+	}
+	return store.ID
+
+}
+
+func GetFirstTimeRun() bool {
+	lock.Lock()
+	defer lock.Unlock()
+	if store.ID == "" {
+		initStore()
+	}
+	return store.FirstTimeRun
+}
+
+func SetFirstTimeRun(val bool) {
+	lock.Lock()
+	defer lock.Unlock()
+	if store.FirstTimeRun == val {
+		return
+	}
+	store.FirstTimeRun = val
+	writeStore(getStorePath())
+}
+
+// lock must be held
+func initStore() {
+	storeFile, err := os.Open(getStorePath())
+	if err == nil {
+		defer storeFile.Close()
+		err = json.NewDecoder(storeFile).Decode(&store)
+		if err == nil {
+			slog.Debug(fmt.Sprintf("loaded existing store %s - ID: %s", getStorePath(), store.ID))
+			return
+		}
+	} else if !errors.Is(err, os.ErrNotExist) {
+		slog.Debug(fmt.Sprintf("unexpected error searching for store: %s", err))
+	}
+	slog.Debug("initializing new store")
+	store.ID = uuid.New().String()
+	writeStore(getStorePath())
+}
+
+func writeStore(storeFilename string) {
+	ollamaDir := filepath.Dir(storeFilename)
+	_, err := os.Stat(ollamaDir)
+	if errors.Is(err, os.ErrNotExist) {
+		if err := os.MkdirAll(ollamaDir, 0o755); err != nil {
+			slog.Error(fmt.Sprintf("create ollama dir %s: %v", ollamaDir, err))
+			return
+		}
+	}
+	payload, err := json.Marshal(store)
+	if err != nil {
+		slog.Error(fmt.Sprintf("failed to marshal store: %s", err))
+		return
+	}
+	fp, err := os.OpenFile(storeFilename, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0o755)
+	if err != nil {
+		slog.Error(fmt.Sprintf("write store payload %s: %v", storeFilename, err))
+		return
+	}
+	defer fp.Close()
+	if n, err := fp.Write(payload); err != nil || n != len(payload) {
+		slog.Error(fmt.Sprintf("write store payload %s: %d vs %d -- %v", storeFilename, n, len(payload), err))
+		return
+	}
+	slog.Debug("Store contents: " + string(payload))
+	slog.Info(fmt.Sprintf("wrote store: %s", storeFilename))
+}

app/store/store_darwin.go

@@ -0,0 +1,13 @@
+package store
+
+import (
+	"os"
+	"path/filepath"
+)
+
+func getStorePath() string {
+	// TODO - system wide location?
+
+	home := os.Getenv("HOME")
+	return filepath.Join(home, "Library", "Application Support", "Ollama", "config.json")
+}

app/store/store_linux.go

@@ -0,0 +1,16 @@
+package store
+
+import (
+	"os"
+	"path/filepath"
+)
+
+func getStorePath() string {
+	if os.Geteuid() == 0 {
+		// TODO where should we store this on linux for system-wide operation?
+		return "/etc/ollama/config.json"
+	}
+
+	home := os.Getenv("HOME")
+	return filepath.Join(home, ".ollama", "config.json")
+}

app/store/store_windows.go

@@ -0,0 +1,11 @@
+package store
+
+import (
+	"os"
+	"path/filepath"
+)
+
+func getStorePath() string {
+	localAppData := os.Getenv("LOCALAPPDATA")
+	return filepath.Join(localAppData, "Ollama", "config.json")
+}

app/tailwind.config.js

@@ -1,6 +0,0 @@
-/** @type {import('tailwindcss').Config} */
-module.exports = {
-  content: ['./src/**/*.{js,ts,jsx,tsx,mdx}'],
-  theme: {},
-  plugins: [],
-}

app/tray/commontray/types.go

@@ -0,0 +1,24 @@
+package commontray
+
+var (
+	Title   = "Ollama"
+	ToolTip = "Ollama"
+
+	UpdateIconName = "tray_upgrade"
+	IconName       = "tray"
+)
+
+type Callbacks struct {
+	Quit       chan struct{}
+	Update     chan struct{}
+	DoFirstUse chan struct{}
+	ShowLogs   chan struct{}
+}
+
+type OllamaTray interface {
+	GetCallbacks() Callbacks
+	Run()
+	UpdateAvailable(ver string) error
+	DisplayFirstUseNotification() error
+	Quit()
+}

app/tray/tray.go

@@ -0,0 +1,28 @@
+package tray
+
+import (
+	"fmt"
+	"runtime"
+
+	"github.com/ollama/ollama/app/assets"
+	"github.com/ollama/ollama/app/tray/commontray"
+)
+
+func NewTray() (commontray.OllamaTray, error) {
+	extension := ".png"
+	if runtime.GOOS == "windows" {
+		extension = ".ico"
+	}
+	iconName := commontray.UpdateIconName + extension
+	updateIcon, err := assets.GetIcon(iconName)
+	if err != nil {
+		return nil, fmt.Errorf("failed to load icon %s: %w", iconName, err)
+	}
+	iconName = commontray.IconName + extension
+	icon, err := assets.GetIcon(iconName)
+	if err != nil {
+		return nil, fmt.Errorf("failed to load icon %s: %w", iconName, err)
+	}
+
+	return InitPlatformTray(icon, updateIcon)
+}

app/tray/tray_darwin.go

@@ -0,0 +1,11 @@
+package tray
+
+import (
+	"fmt"
+
+	"github.com/ollama/ollama/app/tray/commontray"
+)
+
+func InitPlatformTray(icon, updateIcon []byte) (commontray.OllamaTray, error) {
+	return nil, fmt.Errorf("NOT IMPLEMENTED YET")
+}

app/tray/tray_windows.go

@@ -0,0 +1,10 @@
+package tray
+
+import (
+	"github.com/ollama/ollama/app/tray/commontray"
+	"github.com/ollama/ollama/app/tray/wintray"
+)
+
+func InitPlatformTray(icon, updateIcon []byte) (commontray.OllamaTray, error) {
+	return wintray.InitTray(icon, updateIcon)
+}

app/tray/wintray/eventloop.go

@@ -0,0 +1,184 @@
+//go:build windows
+
+package wintray
+
+import (
+	"fmt"
+	"log/slog"
+	"sync"
+	"unsafe"
+
+	"golang.org/x/sys/windows"
+)
+
+var (
+	quitOnce sync.Once
+)
+
+func (t *winTray) Run() {
+	nativeLoop()
+}
+
+func nativeLoop() {
+	// Main message pump.
+	slog.Debug("starting event handling loop")
+	m := &struct {
+		WindowHandle windows.Handle
+		Message      uint32
+		Wparam       uintptr
+		Lparam       uintptr
+		Time         uint32
+		Pt           point
+		LPrivate     uint32
+	}{}
+	for {
+		ret, _, err := pGetMessage.Call(uintptr(unsafe.Pointer(m)), 0, 0, 0)
+
+		// If the function retrieves a message other than WM_QUIT, the return value is nonzero.
+		// If the function retrieves the WM_QUIT message, the return value is zero.
+		// If there is an error, the return value is -1
+		// https://msdn.microsoft.com/en-us/library/windows/desktop/ms644936(v=vs.85).aspx
+		switch int32(ret) {
+		case -1:
+			slog.Error(fmt.Sprintf("get message failure: %v", err))
+			return
+		case 0:
+			return
+		default:
+			pTranslateMessage.Call(uintptr(unsafe.Pointer(m))) //nolint:errcheck
+			pDispatchMessage.Call(uintptr(unsafe.Pointer(m)))  //nolint:errcheck
+
+		}
+	}
+}
+
+// WindowProc callback function that processes messages sent to a window.
+// https://msdn.microsoft.com/en-us/library/windows/desktop/ms633573(v=vs.85).aspx
+func (t *winTray) wndProc(hWnd windows.Handle, message uint32, wParam, lParam uintptr) (lResult uintptr) {
+	const (
+		WM_RBUTTONUP   = 0x0205
+		WM_LBUTTONUP   = 0x0202
+		WM_COMMAND     = 0x0111
+		WM_ENDSESSION  = 0x0016
+		WM_CLOSE       = 0x0010
+		WM_DESTROY     = 0x0002
+		WM_MOUSEMOVE   = 0x0200
+		WM_LBUTTONDOWN = 0x0201
+	)
+	switch message {
+	case WM_COMMAND:
+		menuItemId := int32(wParam)
+		// https://docs.microsoft.com/en-us/windows/win32/menurc/wm-command#menus
+		switch menuItemId {
+		case quitMenuID:
+			select {
+			case t.callbacks.Quit <- struct{}{}:
+			// should not happen but in case not listening
+			default:
+				slog.Error("no listener on Quit")
+			}
+		case updateMenuID:
+			select {
+			case t.callbacks.Update <- struct{}{}:
+			// should not happen but in case not listening
+			default:
+				slog.Error("no listener on Update")
+			}
+		case diagLogsMenuID:
+			select {
+			case t.callbacks.ShowLogs <- struct{}{}:
+			// should not happen but in case not listening
+			default:
+				slog.Error("no listener on ShowLogs")
+			}
+		default:
+			slog.Debug(fmt.Sprintf("Unexpected menu item id: %d", menuItemId))
+		}
+	case WM_CLOSE:
+		boolRet, _, err := pDestroyWindow.Call(uintptr(t.window))
+		if boolRet == 0 {
+			slog.Error(fmt.Sprintf("failed to destroy window: %s", err))
+		}
+		err = t.wcex.unregister()
+		if err != nil {
+			slog.Error(fmt.Sprintf("failed to uregister windo %s", err))
+		}
+	case WM_DESTROY:
+		// same as WM_ENDSESSION, but throws 0 exit code after all
+		defer pPostQuitMessage.Call(uintptr(int32(0))) //nolint:errcheck
+		fallthrough
+	case WM_ENDSESSION:
+		t.muNID.Lock()
+		if t.nid != nil {
+			err := t.nid.delete()
+			if err != nil {
+				slog.Error(fmt.Sprintf("failed to delete nid: %s", err))
+			}
+		}
+		t.muNID.Unlock()
+	case t.wmSystrayMessage:
+		switch lParam {
+		case WM_MOUSEMOVE, WM_LBUTTONDOWN:
+			// Ignore these...
+		case WM_RBUTTONUP, WM_LBUTTONUP:
+			err := t.showMenu()
+			if err != nil {
+				slog.Error(fmt.Sprintf("failed to show menu: %s", err))
+			}
+		case 0x405: // TODO - how is this magic value derived for the notification left click
+			if t.pendingUpdate {
+				select {
+				case t.callbacks.Update <- struct{}{}:
+				// should not happen but in case not listening
+				default:
+					slog.Error("no listener on Update")
+				}
+			} else {
+				select {
+				case t.callbacks.DoFirstUse <- struct{}{}:
+				// should not happen but in case not listening
+				default:
+					slog.Error("no listener on DoFirstUse")
+				}
+			}
+		case 0x404: // Middle click or close notification
+			// slog.Debug("doing nothing on close of first time notification")
+		default:
+			// 0x402 also seems common - what is it?
+			slog.Debug(fmt.Sprintf("unmanaged app message, lParm: 0x%x", lParam))
+		}
+	case t.wmTaskbarCreated: // on explorer.exe restarts
+		t.muNID.Lock()
+		err := t.nid.add()
+		if err != nil {
+			slog.Error(fmt.Sprintf("failed to refresh the taskbar on explorer restart: %s", err))
+		}
+		t.muNID.Unlock()
+	default:
+		// Calls the default window procedure to provide default processing for any window messages that an application does not process.
+		// https://msdn.microsoft.com/en-us/library/windows/desktop/ms633572(v=vs.85).aspx
+		lResult, _, _ = pDefWindowProc.Call(
+			uintptr(hWnd),
+			uintptr(message),
+			uintptr(wParam),
+			uintptr(lParam),
+		)
+	}
+	return
+}
+
+func (t *winTray) Quit() {
+	quitOnce.Do(quit)
+}
+
+func quit() {
+	boolRet, _, err := pPostMessage.Call(
+		uintptr(wt.window),
+		WM_CLOSE,
+		0,
+		0,
+	)
+	if boolRet == 0 {
+		slog.Error(fmt.Sprintf("failed to post close message on shutdown %s", err))
+	}
+}

app/tray/wintray/menus.go

@@ -0,0 +1,71 @@
+//go:build windows
+
+package wintray
+
+import (
+	"fmt"
+	"log/slog"
+	"unsafe"
+
+	"golang.org/x/sys/windows"
+)
+
+const (
+	updatAvailableMenuID = 1
+	updateMenuID         = updatAvailableMenuID + 1
+	separatorMenuID      = updateMenuID + 1
+	diagLogsMenuID       = separatorMenuID + 1
+	diagSeparatorMenuID  = diagLogsMenuID + 1
+	quitMenuID           = diagSeparatorMenuID + 1
+)
+
+func (t *winTray) initMenus() error {
+	if err := t.addOrUpdateMenuItem(diagLogsMenuID, 0, diagLogsMenuTitle, false); err != nil {
+		return fmt.Errorf("unable to create menu entries %w\n", err)
+	}
+	if err := t.addSeparatorMenuItem(diagSeparatorMenuID, 0); err != nil {
+		return fmt.Errorf("unable to create menu entries %w", err)
+	}
+	if err := t.addOrUpdateMenuItem(quitMenuID, 0, quitMenuTitle, false); err != nil {
+		return fmt.Errorf("unable to create menu entries %w\n", err)
+	}
+	return nil
+}
+
+func (t *winTray) UpdateAvailable(ver string) error {
+	if !t.updateNotified {
+		slog.Debug("updating menu and sending notification for new update")
+		if err := t.addOrUpdateMenuItem(updatAvailableMenuID, 0, updateAvailableMenuTitle, true); err != nil {
+			return fmt.Errorf("unable to create menu entries %w", err)
+		}
+		if err := t.addOrUpdateMenuItem(updateMenuID, 0, updateMenutTitle, false); err != nil {
+			return fmt.Errorf("unable to create menu entries %w", err)
+		}
+		if err := t.addSeparatorMenuItem(separatorMenuID, 0); err != nil {
+			return fmt.Errorf("unable to create menu entries %w", err)
+		}
+		iconFilePath, err := iconBytesToFilePath(wt.updateIcon)
+		if err != nil {
+			return fmt.Errorf("unable to write icon data to temp file: %w", err)
+		}
+		if err := wt.setIcon(iconFilePath); err != nil {
+			return fmt.Errorf("unable to set icon: %w", err)
+		}
+		t.updateNotified = true
+
+		t.pendingUpdate = true
+		// Now pop up the notification
+		t.muNID.Lock()
+		defer t.muNID.Unlock()
+		copy(t.nid.InfoTitle[:], windows.StringToUTF16(updateTitle))
+		copy(t.nid.Info[:], windows.StringToUTF16(fmt.Sprintf(updateMessage, ver)))
+		t.nid.Flags |= NIF_INFO
+		t.nid.Timeout = 10
+		t.nid.Size = uint32(unsafe.Sizeof(*wt.nid))
+		err = t.nid.modify()
+		if err != nil {
+			return err
+		}
+	}
+	return nil
+}

app/tray/wintray/messages.go

@@ -0,0 +1,15 @@
+//go:build windows
+
+package wintray
+
+const (
+	firstTimeTitle   = "Ollama is running"
+	firstTimeMessage = "Click here to get started"
+	updateTitle      = "Update available"
+	updateMessage    = "Ollama version %s is ready to install"
+
+	quitMenuTitle            = "Quit Ollama"
+	updateAvailableMenuTitle = "An update is available"
+	updateMenutTitle         = "Restart to update"
+	diagLogsMenuTitle        = "View logs"
+)

app/tray/wintray/notifyicon.go

@@ -0,0 +1,66 @@
+//go:build windows
+
+package wintray
+
+import (
+	"unsafe"
+
+	"golang.org/x/sys/windows"
+)
+
+// Contains information that the system needs to display notifications in the notification area.
+// Used by Shell_NotifyIcon.
+// https://msdn.microsoft.com/en-us/library/windows/desktop/bb773352(v=vs.85).aspx
+// https://msdn.microsoft.com/en-us/library/windows/desktop/bb762159
+type notifyIconData struct {
+	Size                       uint32
+	Wnd                        windows.Handle
+	ID, Flags, CallbackMessage uint32
+	Icon                       windows.Handle
+	Tip                        [128]uint16
+	State, StateMask           uint32
+	Info                       [256]uint16
+	// Timeout, Version           uint32
+	Timeout uint32
+
+	InfoTitle   [64]uint16
+	InfoFlags   uint32
+	GuidItem    windows.GUID
+	BalloonIcon windows.Handle
+}
+
+func (nid *notifyIconData) add() error {
+	const NIM_ADD = 0x00000000
+	res, _, err := pShellNotifyIcon.Call(
+		uintptr(NIM_ADD),
+		uintptr(unsafe.Pointer(nid)),
+	)
+	if res == 0 {
+		return err
+	}
+	return nil
+}
+
+func (nid *notifyIconData) modify() error {
+	const NIM_MODIFY = 0x00000001
+	res, _, err := pShellNotifyIcon.Call(
+		uintptr(NIM_MODIFY),
+		uintptr(unsafe.Pointer(nid)),
+	)
+	if res == 0 {
+		return err
+	}
+	return nil
+}
+
+func (nid *notifyIconData) delete() error {
+	const NIM_DELETE = 0x00000002
+	res, _, err := pShellNotifyIcon.Call(
+		uintptr(NIM_DELETE),
+		uintptr(unsafe.Pointer(nid)),
+	)
+	if res == 0 {
+		return err
+	}
+	return nil
+}

app/tray/wintray/tray.go

@@ -0,0 +1,485 @@
+//go:build windows
+
+package wintray
+
+import (
+	"crypto/md5"
+	"encoding/hex"
+	"fmt"
+	"log/slog"
+	"os"
+	"path/filepath"
+	"sort"
+	"sync"
+	"unsafe"
+
+	"github.com/ollama/ollama/app/tray/commontray"
+	"golang.org/x/sys/windows"
+)
+
+// Helpful sources: https://github.com/golang/exp/blob/master/shiny/driver/internal/win32
+
+// Contains information about loaded resources
+type winTray struct {
+	instance,
+	icon,
+	cursor,
+	window windows.Handle
+
+	loadedImages   map[string]windows.Handle
+	muLoadedImages sync.RWMutex
+
+	// menus keeps track of the submenus keyed by the menu item ID, plus 0
+	// which corresponds to the main popup menu.
+	menus    map[uint32]windows.Handle
+	muMenus  sync.RWMutex
+	menuOf   map[uint32]windows.Handle
+	muMenuOf sync.RWMutex
+	// menuItemIcons maintains the bitmap of each menu item (if applies). It's
+	// needed to show the icon correctly when showing a previously hidden menu
+	// item again.
+	// menuItemIcons   map[uint32]windows.Handle
+	// muMenuItemIcons sync.RWMutex
+	visibleItems   map[uint32][]uint32
+	muVisibleItems sync.RWMutex
+
+	nid   *notifyIconData
+	muNID sync.RWMutex
+	wcex  *wndClassEx
+
+	wmSystrayMessage,
+	wmTaskbarCreated uint32
+
+	pendingUpdate  bool
+	updateNotified bool // Only pop up the notification once - TODO consider daily nag?
+	// Callbacks
+	callbacks  commontray.Callbacks
+	normalIcon []byte
+	updateIcon []byte
+}
+
+var wt winTray
+
+func (t *winTray) GetCallbacks() commontray.Callbacks {
+	return t.callbacks
+}
+
+func InitTray(icon, updateIcon []byte) (*winTray, error) {
+	wt.callbacks.Quit = make(chan struct{})
+	wt.callbacks.Update = make(chan struct{})
+	wt.callbacks.ShowLogs = make(chan struct{})
+	wt.callbacks.DoFirstUse = make(chan struct{})
+	wt.normalIcon = icon
+	wt.updateIcon = updateIcon
+	if err := wt.initInstance(); err != nil {
+		return nil, fmt.Errorf("Unable to init instance: %w\n", err)
+	}
+
+	if err := wt.createMenu(); err != nil {
+		return nil, fmt.Errorf("Unable to create menu: %w\n", err)
+	}
+
+	iconFilePath, err := iconBytesToFilePath(wt.normalIcon)
+	if err != nil {
+		return nil, fmt.Errorf("Unable to write icon data to temp file: %w", err)
+	}
+	if err := wt.setIcon(iconFilePath); err != nil {
+		return nil, fmt.Errorf("Unable to set icon: %w", err)
+	}
+
+	return &wt, wt.initMenus()
+}
+
+func (t *winTray) initInstance() error {
+	const (
+		className  = "OllamaClass"
+		windowName = ""
+	)
+
+	t.wmSystrayMessage = WM_USER + 1
+	t.visibleItems = make(map[uint32][]uint32)
+	t.menus = make(map[uint32]windows.Handle)
+	t.menuOf = make(map[uint32]windows.Handle)
+
+	t.loadedImages = make(map[string]windows.Handle)
+
+	taskbarEventNamePtr, _ := windows.UTF16PtrFromString("TaskbarCreated")
+	// https://msdn.microsoft.com/en-us/library/windows/desktop/ms644947
+	res, _, err := pRegisterWindowMessage.Call(
+		uintptr(unsafe.Pointer(taskbarEventNamePtr)),
+	)
+	if res == 0 { // success 0xc000-0xfff
+		return fmt.Errorf("failed to register window: %w", err)
+	}
+	t.wmTaskbarCreated = uint32(res)
+
+	instanceHandle, _, err := pGetModuleHandle.Call(0)
+	if instanceHandle == 0 {
+		return err
+	}
+	t.instance = windows.Handle(instanceHandle)
+
+	// https://msdn.microsoft.com/en-us/library/windows/desktop/ms648072(v=vs.85).aspx
+	iconHandle, _, err := pLoadIcon.Call(0, uintptr(IDI_APPLICATION))
+	if iconHandle == 0 {
+		return err
+	}
+	t.icon = windows.Handle(iconHandle)
+
+	// https://msdn.microsoft.com/en-us/library/windows/desktop/ms648391(v=vs.85).aspx
+	cursorHandle, _, err := pLoadCursor.Call(0, uintptr(IDC_ARROW))
+	if cursorHandle == 0 {
+		return err
+	}
+	t.cursor = windows.Handle(cursorHandle)
+
+	classNamePtr, err := windows.UTF16PtrFromString(className)
+	if err != nil {
+		return err
+	}
+
+	windowNamePtr, err := windows.UTF16PtrFromString(windowName)
+	if err != nil {
+		return err
+	}
+
+	t.wcex = &wndClassEx{
+		Style:      CS_HREDRAW | CS_VREDRAW,
+		WndProc:    windows.NewCallback(t.wndProc),
+		Instance:   t.instance,
+		Icon:       t.icon,
+		Cursor:     t.cursor,
+		Background: windows.Handle(6), // (COLOR_WINDOW + 1)
+		ClassName:  classNamePtr,
+		IconSm:     t.icon,
+	}
+	if err := t.wcex.register(); err != nil {
+		return err
+	}
+
+	windowHandle, _, err := pCreateWindowEx.Call(
+		uintptr(0),
+		uintptr(unsafe.Pointer(classNamePtr)),
+		uintptr(unsafe.Pointer(windowNamePtr)),
+		uintptr(WS_OVERLAPPEDWINDOW),
+		uintptr(CW_USEDEFAULT),
+		uintptr(CW_USEDEFAULT),
+		uintptr(CW_USEDEFAULT),
+		uintptr(CW_USEDEFAULT),
+		uintptr(0),
+		uintptr(0),
+		uintptr(t.instance),
+		uintptr(0),
+	)
+	if windowHandle == 0 {
+		return err
+	}
+	t.window = windows.Handle(windowHandle)
+
+	pShowWindow.Call(uintptr(t.window), uintptr(SW_HIDE)) //nolint:errcheck
+
+	boolRet, _, err := pUpdateWindow.Call(uintptr(t.window))
+	if boolRet == 0 {
+		slog.Error(fmt.Sprintf("failed to update window: %s", err))
+	}
+
+	t.muNID.Lock()
+	defer t.muNID.Unlock()
+	t.nid = &notifyIconData{
+		Wnd:             windows.Handle(t.window),
+		ID:              100,
+		Flags:           NIF_MESSAGE,
+		CallbackMessage: t.wmSystrayMessage,
+	}
+	t.nid.Size = uint32(unsafe.Sizeof(*t.nid))
+
+	return t.nid.add()
+}
+
+func (t *winTray) createMenu() error {
+
+	menuHandle, _, err := pCreatePopupMenu.Call()
+	if menuHandle == 0 {
+		return err
+	}
+	t.menus[0] = windows.Handle(menuHandle)
+
+	// https://msdn.microsoft.com/en-us/library/windows/desktop/ms647575(v=vs.85).aspx
+	mi := struct {
+		Size, Mask, Style, Max uint32
+		Background             windows.Handle
+		ContextHelpID          uint32
+		MenuData               uintptr
+	}{
+		Mask: MIM_APPLYTOSUBMENUS,
+	}
+	mi.Size = uint32(unsafe.Sizeof(mi))
+
+	res, _, err := pSetMenuInfo.Call(
+		uintptr(t.menus[0]),
+		uintptr(unsafe.Pointer(&mi)),
+	)
+	if res == 0 {
+		return err
+	}
+	return nil
+}
+
+// Contains information about a menu item.
+// https://msdn.microsoft.com/en-us/library/windows/desktop/ms647578(v=vs.85).aspx
+type menuItemInfo struct {
+	Size, Mask, Type, State     uint32
+	ID                          uint32
+	SubMenu, Checked, Unchecked windows.Handle
+	ItemData                    uintptr
+	TypeData                    *uint16
+	Cch                         uint32
+	BMPItem                     windows.Handle
+}
+
+func (t *winTray) addOrUpdateMenuItem(menuItemId uint32, parentId uint32, title string, disabled bool) error {
+	titlePtr, err := windows.UTF16PtrFromString(title)
+	if err != nil {
+		return err
+	}
+
+	mi := menuItemInfo{
+		Mask:     MIIM_FTYPE | MIIM_STRING | MIIM_ID | MIIM_STATE,
+		Type:     MFT_STRING,
+		ID:       uint32(menuItemId),
+		TypeData: titlePtr,
+		Cch:      uint32(len(title)),
+	}
+	mi.Size = uint32(unsafe.Sizeof(mi))
+	if disabled {
+		mi.State |= MFS_DISABLED
+	}
+
+	var res uintptr
+	t.muMenus.RLock()
+	menu := t.menus[parentId]
+	t.muMenus.RUnlock()
+	if t.getVisibleItemIndex(parentId, menuItemId) != -1 {
+		// We set the menu item info based on the menuID
+		boolRet, _, err := pSetMenuItemInfo.Call(
+			uintptr(menu),
+			uintptr(menuItemId),
+			0,
+			uintptr(unsafe.Pointer(&mi)),
+		)
+		if boolRet == 0 {
+			return fmt.Errorf("failed to set menu item: %w", err)
+		}
+	}
+
+	if res == 0 {
+		// Menu item does not already exist, create it
+		t.muMenus.RLock()
+		submenu, exists := t.menus[menuItemId]
+		t.muMenus.RUnlock()
+		if exists {
+			mi.Mask |= MIIM_SUBMENU
+			mi.SubMenu = submenu
+		}
+		t.addToVisibleItems(parentId, menuItemId)
+		position := t.getVisibleItemIndex(parentId, menuItemId)
+		res, _, err = pInsertMenuItem.Call(
+			uintptr(menu),
+			uintptr(position),
+			1,
+			uintptr(unsafe.Pointer(&mi)),
+		)
+		if res == 0 {
+			t.delFromVisibleItems(parentId, menuItemId)
+			return err
+		}
+		t.muMenuOf.Lock()
+		t.menuOf[menuItemId] = menu
+		t.muMenuOf.Unlock()
+	}
+
+	return nil
+}
+
+func (t *winTray) addSeparatorMenuItem(menuItemId, parentId uint32) error {
+
+	mi := menuItemInfo{
+		Mask: MIIM_FTYPE | MIIM_ID | MIIM_STATE,
+		Type: MFT_SEPARATOR,
+		ID:   uint32(menuItemId),
+	}
+
+	mi.Size = uint32(unsafe.Sizeof(mi))
+
+	t.addToVisibleItems(parentId, menuItemId)
+	position := t.getVisibleItemIndex(parentId, menuItemId)
+	t.muMenus.RLock()
+	menu := uintptr(t.menus[parentId])
+	t.muMenus.RUnlock()
+	res, _, err := pInsertMenuItem.Call(
+		menu,
+		uintptr(position),
+		1,
+		uintptr(unsafe.Pointer(&mi)),
+	)
+	if res == 0 {
+		return err
+	}
+
+	return nil
+}
+
+// func (t *winTray) hideMenuItem(menuItemId, parentId uint32) error {
+// 	const ERROR_SUCCESS syscall.Errno = 0
+
+// 	t.muMenus.RLock()
+// 	menu := uintptr(t.menus[parentId])
+// 	t.muMenus.RUnlock()
+// 	res, _, err := pRemoveMenu.Call(
+// 		menu,
+// 		uintptr(menuItemId),
+// 		MF_BYCOMMAND,
+// 	)
+// 	if res == 0 && err.(syscall.Errno) != ERROR_SUCCESS {
+// 		return err
+// 	}
+// 	t.delFromVisibleItems(parentId, menuItemId)
+
+// 	return nil
+// }
+
+func (t *winTray) showMenu() error {
+	p := point{}
+	boolRet, _, err := pGetCursorPos.Call(uintptr(unsafe.Pointer(&p)))
+	if boolRet == 0 {
+		return err
+	}
+	boolRet, _, err = pSetForegroundWindow.Call(uintptr(t.window))
+	if boolRet == 0 {
+		slog.Warn(fmt.Sprintf("failed to bring menu to foreground: %s", err))
+	}
+
+	boolRet, _, err = pTrackPopupMenu.Call(
+		uintptr(t.menus[0]),
+		TPM_BOTTOMALIGN|TPM_LEFTALIGN,
+		uintptr(p.X),
+		uintptr(p.Y),
+		0,
+		uintptr(t.window),
+		0,
+	)
+	if boolRet == 0 {
+		return err
+	}
+
+	return nil
+}
+
+func (t *winTray) delFromVisibleItems(parent, val uint32) {
+	t.muVisibleItems.Lock()
+	defer t.muVisibleItems.Unlock()
+	visibleItems := t.visibleItems[parent]
+	for i, itemval := range visibleItems {
+		if val == itemval {
+			t.visibleItems[parent] = append(visibleItems[:i], visibleItems[i+1:]...)
+			break
+		}
+	}
+}
+
+func (t *winTray) addToVisibleItems(parent, val uint32) {
+	t.muVisibleItems.Lock()
+	defer t.muVisibleItems.Unlock()
+	if visibleItems, exists := t.visibleItems[parent]; !exists {
+		t.visibleItems[parent] = []uint32{val}
+	} else {
+		newvisible := append(visibleItems, val)
+		sort.Slice(newvisible, func(i, j int) bool { return newvisible[i] < newvisible[j] })
+		t.visibleItems[parent] = newvisible
+	}
+}
+
+func (t *winTray) getVisibleItemIndex(parent, val uint32) int {
+	t.muVisibleItems.RLock()
+	defer t.muVisibleItems.RUnlock()
+	for i, itemval := range t.visibleItems[parent] {
+		if val == itemval {
+			return i
+		}
+	}
+	return -1
+}
+
+func iconBytesToFilePath(iconBytes []byte) (string, error) {
+	bh := md5.Sum(iconBytes)
+	dataHash := hex.EncodeToString(bh[:])
+	iconFilePath := filepath.Join(os.TempDir(), "ollama_temp_icon_"+dataHash)
+
+	if _, err := os.Stat(iconFilePath); os.IsNotExist(err) {
+		if err := os.WriteFile(iconFilePath, iconBytes, 0644); err != nil {
+			return "", err
+		}
+	}
+	return iconFilePath, nil
+}
+
+// Loads an image from file and shows it in tray.
+// Shell_NotifyIcon: https://msdn.microsoft.com/en-us/library/windows/desktop/bb762159(v=vs.85).aspx
+func (t *winTray) setIcon(src string) error {
+
+	h, err := t.loadIconFrom(src)
+	if err != nil {
+		return err
+	}
+
+	t.muNID.Lock()
+	defer t.muNID.Unlock()
+	t.nid.Icon = h
+	t.nid.Flags |= NIF_ICON
+	t.nid.Size = uint32(unsafe.Sizeof(*t.nid))
+
+	return t.nid.modify()
+}
+
+// Loads an image from file to be shown in tray or menu item.
+// LoadImage: https://msdn.microsoft.com/en-us/library/windows/desktop/ms648045(v=vs.85).aspx
+func (t *winTray) loadIconFrom(src string) (windows.Handle, error) {
+
+	// Save and reuse handles of loaded images
+	t.muLoadedImages.RLock()
+	h, ok := t.loadedImages[src]
+	t.muLoadedImages.RUnlock()
+	if !ok {
+		srcPtr, err := windows.UTF16PtrFromString(src)
+		if err != nil {
+			return 0, err
+		}
+		res, _, err := pLoadImage.Call(
+			0,
+			uintptr(unsafe.Pointer(srcPtr)),
+			IMAGE_ICON,
+			0,
+			0,
+			LR_LOADFROMFILE|LR_DEFAULTSIZE,
+		)
+		if res == 0 {
+			return 0, err
+		}
+		h = windows.Handle(res)
+		t.muLoadedImages.Lock()
+		t.loadedImages[src] = h
+		t.muLoadedImages.Unlock()
+	}
+	return h, nil
+}
+
+func (t *winTray) DisplayFirstUseNotification() error {
+	t.muNID.Lock()
+	defer t.muNID.Unlock()
+	copy(t.nid.InfoTitle[:], windows.StringToUTF16(firstTimeTitle))
+	copy(t.nid.Info[:], windows.StringToUTF16(firstTimeMessage))
+	t.nid.Flags |= NIF_INFO
+	t.nid.Size = uint32(unsafe.Sizeof(*wt.nid))
+
+	return t.nid.modify()
+}

app/tray/wintray/w32api.go

@@ -0,0 +1,89 @@
+//go:build windows
+
+package wintray
+
+import (
+	"runtime"
+
+	"golang.org/x/sys/windows"
+)
+
+var (
+	k32 = windows.NewLazySystemDLL("Kernel32.dll")
+	u32 = windows.NewLazySystemDLL("User32.dll")
+	s32 = windows.NewLazySystemDLL("Shell32.dll")
+
+	pCreatePopupMenu       = u32.NewProc("CreatePopupMenu")
+	pCreateWindowEx        = u32.NewProc("CreateWindowExW")
+	pDefWindowProc         = u32.NewProc("DefWindowProcW")
+	pDestroyWindow         = u32.NewProc("DestroyWindow")
+	pDispatchMessage       = u32.NewProc("DispatchMessageW")
+	pGetCursorPos          = u32.NewProc("GetCursorPos")
+	pGetMessage            = u32.NewProc("GetMessageW")
+	pGetModuleHandle       = k32.NewProc("GetModuleHandleW")
+	pInsertMenuItem        = u32.NewProc("InsertMenuItemW")
+	pLoadCursor            = u32.NewProc("LoadCursorW")
+	pLoadIcon              = u32.NewProc("LoadIconW")
+	pLoadImage             = u32.NewProc("LoadImageW")
+	pPostMessage           = u32.NewProc("PostMessageW")
+	pPostQuitMessage       = u32.NewProc("PostQuitMessage")
+	pRegisterClass         = u32.NewProc("RegisterClassExW")
+	pRegisterWindowMessage = u32.NewProc("RegisterWindowMessageW")
+	pSetForegroundWindow   = u32.NewProc("SetForegroundWindow")
+	pSetMenuInfo           = u32.NewProc("SetMenuInfo")
+	pSetMenuItemInfo       = u32.NewProc("SetMenuItemInfoW")
+	pShellNotifyIcon       = s32.NewProc("Shell_NotifyIconW")
+	pShowWindow            = u32.NewProc("ShowWindow")
+	pTrackPopupMenu        = u32.NewProc("TrackPopupMenu")
+	pTranslateMessage      = u32.NewProc("TranslateMessage")
+	pUnregisterClass       = u32.NewProc("UnregisterClassW")
+	pUpdateWindow          = u32.NewProc("UpdateWindow")
+)
+
+const (
+	CS_HREDRAW          = 0x0002
+	CS_VREDRAW          = 0x0001
+	CW_USEDEFAULT       = 0x80000000
+	IDC_ARROW           = 32512 // Standard arrow
+	IDI_APPLICATION     = 32512
+	IMAGE_ICON          = 1          // Loads an icon
+	LR_DEFAULTSIZE      = 0x00000040 // Loads default-size icon for windows(SM_CXICON x SM_CYICON) if cx, cy are set to zero
+	LR_LOADFROMFILE     = 0x00000010 // Loads the stand-alone image from the file
+	MF_BYCOMMAND        = 0x00000000
+	MFS_DISABLED        = 0x00000003
+	MFT_SEPARATOR       = 0x00000800
+	MFT_STRING          = 0x00000000
+	MIIM_BITMAP         = 0x00000080
+	MIIM_FTYPE          = 0x00000100
+	MIIM_ID             = 0x00000002
+	MIIM_STATE          = 0x00000001
+	MIIM_STRING         = 0x00000040
+	MIIM_SUBMENU        = 0x00000004
+	MIM_APPLYTOSUBMENUS = 0x80000000
+	NIF_ICON            = 0x00000002
+	NIF_INFO            = 0x00000010
+	NIF_MESSAGE         = 0x00000001
+	SW_HIDE             = 0
+	TPM_BOTTOMALIGN     = 0x0020
+	TPM_LEFTALIGN       = 0x0000
+	WM_CLOSE            = 0x0010
+	WM_USER             = 0x0400
+	WS_CAPTION          = 0x00C00000
+	WS_MAXIMIZEBOX      = 0x00010000
+	WS_MINIMIZEBOX      = 0x00020000
+	WS_OVERLAPPED       = 0x00000000
+	WS_OVERLAPPEDWINDOW = WS_OVERLAPPED | WS_CAPTION | WS_SYSMENU | WS_THICKFRAME | WS_MINIMIZEBOX | WS_MAXIMIZEBOX
+	WS_SYSMENU          = 0x00080000
+	WS_THICKFRAME       = 0x00040000
+)
+
+// Not sure if this is actually needed on windows
+func init() {
+	runtime.LockOSThread()
+}
+
+// The POINT structure defines the x- and y- coordinates of a point.
+// https://msdn.microsoft.com/en-us/library/windows/desktop/dd162805(v=vs.85).aspx
+type point struct {
+	X, Y int32
+}

app/tray/wintray/winclass.go

@@ -0,0 +1,45 @@
+//go:build windows
+
+package wintray
+
+import (
+	"unsafe"
+
+	"golang.org/x/sys/windows"
+)
+
+// Contains window class information.
+// It is used with the RegisterClassEx and GetClassInfoEx functions.
+// https://msdn.microsoft.com/en-us/library/ms633577.aspx
+type wndClassEx struct {
+	Size, Style                        uint32
+	WndProc                            uintptr
+	ClsExtra, WndExtra                 int32
+	Instance, Icon, Cursor, Background windows.Handle
+	MenuName, ClassName                *uint16
+	IconSm                             windows.Handle
+}
+
+// Registers a window class for subsequent use in calls to the CreateWindow or CreateWindowEx function.
+// https://msdn.microsoft.com/en-us/library/ms633587.aspx
+func (w *wndClassEx) register() error {
+	w.Size = uint32(unsafe.Sizeof(*w))
+	res, _, err := pRegisterClass.Call(uintptr(unsafe.Pointer(w)))
+	if res == 0 {
+		return err
+	}
+	return nil
+}
+
+// Unregisters a window class, freeing the memory required for the class.
+// https://msdn.microsoft.com/en-us/library/ms644899.aspx
+func (w *wndClassEx) unregister() error {
+	res, _, err := pUnregisterClass.Call(
+		uintptr(unsafe.Pointer(w.ClassName)),
+		uintptr(w.Instance),
+	)
+	if res == 0 {
+		return err
+	}
+	return nil
+}

app/tsconfig.json

@@ -1,20 +0,0 @@
-{
-  "compilerOptions": {
-    "target": "ES6",
-    "allowJs": true,
-    "module": "commonjs",
-    "skipLibCheck": true,
-    "esModuleInterop": true,
-    "noImplicitAny": true,
-    "sourceMap": true,
-    "baseUrl": ".",
-    "outDir": "dist",
-    "moduleResolution": "node",
-    "resolveJsonModule": true,
-    "paths": {
-      "*": ["node_modules/*"]
-    },
-    "jsx": "react-jsx"
-  },
-  "include": ["src/**/*"]
-}

app/updater/updater.go

@@ -0,0 +1,232 @@
+package updater
+
+import (
+	"context"
+	"crypto/rand"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"log/slog"
+	"mime"
+	"net/http"
+	"net/url"
+	"os"
+	"path"
+	"path/filepath"
+	"runtime"
+	"strings"
+	"time"
+
+	"github.com/ollama/ollama/auth"
+	"github.com/ollama/ollama/version"
+)
+
+var (
+	UpdateStageDir string
+)
+
+var (
+	UpdateCheckURLBase  = "https://ollama.com/api/update"
+	UpdateDownloaded    = false
+	UpdateCheckInterval = 60 * 60 * time.Second
+)
+
+// TODO - maybe move up to the API package?
+type UpdateResponse struct {
+	UpdateURL     string `json:"url"`
+	UpdateVersion string `json:"version"`
+}
+
+func IsNewReleaseAvailable(ctx context.Context) (bool, UpdateResponse) {
+	var updateResp UpdateResponse
+
+	requestURL, err := url.Parse(UpdateCheckURLBase)
+	if err != nil {
+		return false, updateResp
+	}
+
+	query := requestURL.Query()
+	query.Add("os", runtime.GOOS)
+	query.Add("arch", runtime.GOARCH)
+	query.Add("version", version.Version)
+	query.Add("ts", fmt.Sprintf("%d", time.Now().Unix()))
+
+	nonce, err := auth.NewNonce(rand.Reader, 16)
+	if err != nil {
+		return false, updateResp
+	}
+
+	query.Add("nonce", nonce)
+	requestURL.RawQuery = query.Encode()
+
+	data := []byte(fmt.Sprintf("%s,%s", http.MethodGet, requestURL.RequestURI()))
+	signature, err := auth.Sign(ctx, data)
+	if err != nil {
+		return false, updateResp
+	}
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, requestURL.String(), nil)
+	if err != nil {
+		slog.Warn(fmt.Sprintf("failed to check for update: %s", err))
+		return false, updateResp
+	}
+	req.Header.Set("Authorization", signature)
+	req.Header.Set("User-Agent", fmt.Sprintf("ollama/%s (%s %s) Go/%s", version.Version, runtime.GOARCH, runtime.GOOS, runtime.Version()))
+
+	slog.Debug("checking for available update", "requestURL", requestURL)
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		slog.Warn(fmt.Sprintf("failed to check for update: %s", err))
+		return false, updateResp
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode == 204 {
+		slog.Debug("check update response 204 (current version is up to date)")
+		return false, updateResp
+	}
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		slog.Warn(fmt.Sprintf("failed to read body response: %s", err))
+	}
+
+	if resp.StatusCode != 200 {
+		slog.Info(fmt.Sprintf("check update error %d - %.96s", resp.StatusCode, string(body)))
+		return false, updateResp
+	}
+	err = json.Unmarshal(body, &updateResp)
+	if err != nil {
+		slog.Warn(fmt.Sprintf("malformed response checking for update: %s", err))
+		return false, updateResp
+	}
+	// Extract the version string from the URL in the github release artifact path
+	updateResp.UpdateVersion = path.Base(path.Dir(updateResp.UpdateURL))
+
+	slog.Info("New update available at " + updateResp.UpdateURL)
+	return true, updateResp
+}
+
+func DownloadNewRelease(ctx context.Context, updateResp UpdateResponse) error {
+	// Do a head first to check etag info
+	req, err := http.NewRequestWithContext(ctx, http.MethodHead, updateResp.UpdateURL, nil)
+	if err != nil {
+		return err
+	}
+
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		return fmt.Errorf("error checking update: %w", err)
+	}
+	if resp.StatusCode != 200 {
+		return fmt.Errorf("unexpected status attempting to download update %d", resp.StatusCode)
+	}
+	resp.Body.Close()
+	etag := strings.Trim(resp.Header.Get("etag"), "\"")
+	if etag == "" {
+		slog.Debug("no etag detected, falling back to filename based dedup")
+		etag = "_"
+	}
+	filename := "OllamaSetup.exe"
+	_, params, err := mime.ParseMediaType(resp.Header.Get("content-disposition"))
+	if err == nil {
+		filename = params["filename"]
+	}
+
+	stageFilename := filepath.Join(UpdateStageDir, etag, filename)
+
+	// Check to see if we already have it downloaded
+	_, err = os.Stat(stageFilename)
+	if err == nil {
+		slog.Info("update already downloaded")
+		return nil
+	}
+
+	cleanupOldDownloads()
+
+	req.Method = http.MethodGet
+	resp, err = http.DefaultClient.Do(req)
+	if err != nil {
+		return fmt.Errorf("error checking update: %w", err)
+	}
+	defer resp.Body.Close()
+	etag = strings.Trim(resp.Header.Get("etag"), "\"")
+	if etag == "" {
+		slog.Debug("no etag detected, falling back to filename based dedup") // TODO probably can get rid of this redundant log
+		etag = "_"
+	}
+
+	stageFilename = filepath.Join(UpdateStageDir, etag, filename)
+
+	_, err = os.Stat(filepath.Dir(stageFilename))
+	if errors.Is(err, os.ErrNotExist) {
+		if err := os.MkdirAll(filepath.Dir(stageFilename), 0o755); err != nil {
+			return fmt.Errorf("create ollama dir %s: %v", filepath.Dir(stageFilename), err)
+		}
+	}
+
+	payload, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return fmt.Errorf("failed to read body response: %w", err)
+	}
+	fp, err := os.OpenFile(stageFilename, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0o755)
+	if err != nil {
+		return fmt.Errorf("write payload %s: %w", stageFilename, err)
+	}
+	defer fp.Close()
+	if n, err := fp.Write(payload); err != nil || n != len(payload) {
+		return fmt.Errorf("write payload %s: %d vs %d -- %w", stageFilename, n, len(payload), err)
+	}
+	slog.Info("new update downloaded " + stageFilename)
+
+	UpdateDownloaded = true
+	return nil
+}
+
+func cleanupOldDownloads() {
+	files, err := os.ReadDir(UpdateStageDir)
+	if err != nil && errors.Is(err, os.ErrNotExist) {
+		// Expected behavior on first run
+		return
+	} else if err != nil {
+		slog.Warn(fmt.Sprintf("failed to list stage dir: %s", err))
+		return
+	}
+	for _, file := range files {
+		fullname := filepath.Join(UpdateStageDir, file.Name())
+		slog.Debug("cleaning up old download: " + fullname)
+		err = os.RemoveAll(fullname)
+		if err != nil {
+			slog.Warn(fmt.Sprintf("failed to cleanup stale update download %s", err))
+		}
+	}
+}
+
+func StartBackgroundUpdaterChecker(ctx context.Context, cb func(string) error) {
+	go func() {
+		// Don't blast an update message immediately after startup
+		// time.Sleep(30 * time.Second)
+		time.Sleep(3 * time.Second)
+
+		for {
+			available, resp := IsNewReleaseAvailable(ctx)
+			if available {
+				err := DownloadNewRelease(ctx, resp)
+				if err != nil {
+					slog.Error(fmt.Sprintf("failed to download new release: %s", err))
+				}
+				err = cb(resp.UpdateVersion)
+				if err != nil {
+					slog.Warn(fmt.Sprintf("failed to register update available with tray: %s", err))
+				}
+			}
+			select {
+			case <-ctx.Done():
+				slog.Debug("stopping background update checker")
+				return
+			default:
+				time.Sleep(UpdateCheckInterval)
+			}
+		}
+	}()
+}

app/updater/updater_darwin.go

@@ -0,0 +1,10 @@
+package updater
+
+import (
+	"context"
+	"fmt"
+)
+
+func DoUpgrade(cancel context.CancelFunc, done chan int) error {
+	return fmt.Errorf("DoUpgrade not yet implemented")
+}

app/updater/updater_windows.go

@@ -0,0 +1,86 @@
+package updater
+
+import (
+	"context"
+	"fmt"
+	"log/slog"
+	"os"
+	"os/exec"
+	"path/filepath"
+)
+
+func init() {
+	UpdateStageDir = filepath.Join(os.Getenv("LOCALAPPDATA"), "Ollama", "updates")
+}
+
+func DoUpgrade(cancel context.CancelFunc, done chan int) error {
+	logFile := filepath.Join(os.Getenv("LOCALAPPDATA"), "Ollama", "upgrade.log")
+
+	files, err := filepath.Glob(filepath.Join(UpdateStageDir, "*", "*.exe")) // TODO generalize for multiplatform
+	if err != nil {
+		return fmt.Errorf("failed to lookup downloads: %s", err)
+	}
+	if len(files) == 0 {
+		return fmt.Errorf("no update downloads found")
+	} else if len(files) > 1 {
+		// Shouldn't happen
+		slog.Warn(fmt.Sprintf("multiple downloads found, using first one %v", files))
+	}
+	installerExe := files[0]
+
+	slog.Info("starting upgrade with " + installerExe)
+	slog.Info("upgrade log file " + logFile)
+
+	// When running in debug mode, we'll be "verbose" and let the installer pop up and prompt
+	installArgs := []string{
+		"/CLOSEAPPLICATIONS",             // Quit the tray app if it's still running
+		"/LOG=" + filepath.Base(logFile), // Only relative seems reliable, so set pwd
+		"/FORCECLOSEAPPLICATIONS",        // Force close the tray app - might be needed
+	}
+	// When we're not in debug mode, make the upgrade as quiet as possible (no GUI, no prompts)
+	// TODO - temporarily disable since we're pinning in debug mode for the preview
+	// if debug := os.Getenv("OLLAMA_DEBUG"); debug == "" {
+	installArgs = append(installArgs,
+		"/SP", // Skip the "This will install... Do you wish to continue" prompt
+		"/SUPPRESSMSGBOXES",
+		"/SILENT",
+		"/VERYSILENT",
+	)
+	// }
+
+	// Safeguard in case we have requests in flight that need to drain...
+	slog.Info("Waiting for server to shutdown")
+	cancel()
+	if done != nil {
+		<-done
+	} else {
+		// Shouldn't happen
+		slog.Warn("done chan was nil, not actually waiting")
+	}
+
+	slog.Debug(fmt.Sprintf("starting installer: %s %v", installerExe, installArgs))
+	os.Chdir(filepath.Dir(logFile)) //nolint:errcheck
+	cmd := exec.Command(installerExe, installArgs...)
+
+	if err := cmd.Start(); err != nil {
+		return fmt.Errorf("unable to start ollama app %w", err)
+	}
+
+	if cmd.Process != nil {
+		err = cmd.Process.Release()
+		if err != nil {
+			slog.Error(fmt.Sprintf("failed to release server process: %s", err))
+		}
+	} else {
+		// TODO - some details about why it didn't start, or is this a pedantic error case?
+		return fmt.Errorf("installer process did not start")
+	}
+
+	// TODO should we linger for a moment and check to make sure it's actually running by checking the pid?
+
+	slog.Info("Installer started in background, exiting")
+
+	os.Exit(0)
+	// Not reached
+	return nil
+}

app/webpack.main.config.ts

@@ -1,20 +0,0 @@
-import type { Configuration } from 'webpack'
-
-import { rules } from './webpack.rules'
-import { plugins } from './webpack.plugins'
-
-export const mainConfig: Configuration = {
-  /**
-   * This is the main entry point for your application, it's the first file
-   * that runs in the main process.
-   */
-  entry: './src/index.ts',
-  // Put your normal webpack config below here
-  module: {
-    rules,
-  },
-  plugins,
-  resolve: {
-    extensions: ['.js', '.ts', '.jsx', '.tsx', '.css', '.json'],
-  },
-}

app/webpack.plugins.ts

@@ -1,14 +0,0 @@
-import type IForkTsCheckerWebpackPlugin from 'fork-ts-checker-webpack-plugin'
-import { DefinePlugin } from 'webpack'
-
-// eslint-disable-next-line @typescript-eslint/no-var-requires
-const ForkTsCheckerWebpackPlugin: typeof IForkTsCheckerWebpackPlugin = require('fork-ts-checker-webpack-plugin')
-
-export const plugins = [
-  new ForkTsCheckerWebpackPlugin({
-    logger: 'webpack-infrastructure',
-  }),
-  new DefinePlugin({
-    'process.env.TELEMETRY_WRITE_KEY': JSON.stringify(process.env.TELEMETRY_WRITE_KEY),
-  }),
-]

app/webpack.renderer.config.ts

@@ -1,19 +0,0 @@
-import type { Configuration } from 'webpack'
-
-import { rules } from './webpack.rules'
-import { plugins } from './webpack.plugins'
-
-rules.push({
-  test: /\.css$/,
-  use: [{ loader: 'style-loader' }, { loader: 'css-loader' }, { loader: 'postcss-loader' }],
-})
-
-export const rendererConfig: Configuration = {
-  module: {
-    rules,
-  },
-  plugins,
-  resolve: {
-    extensions: ['.js', '.ts', '.jsx', '.tsx', '.css'],
-  },
-}

app/webpack.rules.ts

@@ -1,35 +0,0 @@
-import type { ModuleOptions } from 'webpack'
-
-export const rules: Required<ModuleOptions>['rules'] = [
-  // Add support for native node modules
-  {
-    // We're specifying native_modules in the test because the asset relocator loader generates a
-    // "fake" .node file which is really a cjs file.
-    test: /native_modules[/\\].+\.node$/,
-    use: 'node-loader',
-  },
-  {
-    test: /[/\\]node_modules[/\\].+\.(m?js|node)$/,
-    parser: { amd: false },
-    use: {
-      loader: '@vercel/webpack-asset-relocator-loader',
-      options: {
-        outputAssetBase: 'native_modules',
-      },
-    },
-  },
-  {
-    test: /\.tsx?$/,
-    exclude: /(node_modules|\.webpack)/,
-    use: {
-      loader: 'ts-loader',
-      options: {
-        transpileOnly: true,
-      },
-    },
-  },
-  {
-    test: /\.svg$/,
-    use: ['@svgr/webpack'],
-  },
-]

app/windows/ollama.iss

@@ -0,0 +1,156 @@
+; Inno Setup Installer for Ollama
+;
+; To build the installer use the build script invoked from the top of the source tree
+; 
+; powershell -ExecutionPolicy Bypass -File .\scripts\build_windows.ps
+
+
+#define MyAppName "Ollama"
+#if GetEnv("PKG_VERSION") != ""
+  #define MyAppVersion GetEnv("PKG_VERSION")
+#else
+  #define MyAppVersion "0.0.0"
+#endif
+#define MyAppPublisher "Ollama"
+#define MyAppURL "https://ollama.com/"
+#define MyAppExeName "ollama app.exe"
+#define MyIcon ".\assets\app.ico"
+
+[Setup]
+; NOTE: The value of AppId uniquely identifies this application. Do not use the same AppId value in installers for other applications.
+; (To generate a new GUID, click Tools | Generate GUID inside the IDE.)
+AppId={{44E83376-CE68-45EB-8FC1-393500EB558C}
+AppName={#MyAppName}
+AppVersion={#MyAppVersion}
+VersionInfoVersion={#MyAppVersion}
+;AppVerName={#MyAppName} {#MyAppVersion}
+AppPublisher={#MyAppPublisher}
+AppPublisherURL={#MyAppURL}
+AppSupportURL={#MyAppURL}
+AppUpdatesURL={#MyAppURL}
+ArchitecturesAllowed=x64 arm64
+ArchitecturesInstallIn64BitMode=x64 arm64
+DefaultDirName={localappdata}\Programs\{#MyAppName}
+DefaultGroupName={#MyAppName}
+DisableProgramGroupPage=yes
+PrivilegesRequired=lowest
+OutputBaseFilename="OllamaSetup"
+SetupIconFile={#MyIcon}
+UninstallDisplayIcon={uninstallexe}
+Compression=lzma2
+SolidCompression=no
+WizardStyle=modern
+ChangesEnvironment=yes
+OutputDir=..\dist\
+
+; Disable logging once everything's battle tested
+; Filename will be %TEMP%\Setup Log*.txt
+SetupLogging=yes
+CloseApplications=yes
+RestartApplications=no
+
+; https://jrsoftware.org/ishelp/index.php?topic=setup_wizardimagefile
+WizardSmallImageFile=.\assets\setup.bmp
+
+; TODO verifty actual min windows version...
+; OG Win 10
+MinVersion=10.0.10240
+
+; First release that supports WinRT UI Composition for win32 apps
+; MinVersion=10.0.17134
+; First release with XAML Islands - possible UI path forward
+; MinVersion=10.0.18362
+
+; quiet...
+DisableDirPage=yes
+DisableFinishedPage=yes
+DisableReadyMemo=yes
+DisableReadyPage=yes
+DisableStartupPrompt=yes
+DisableWelcomePage=yes
+
+; TODO - percentage can't be set less than 100, so how to make it shorter?
+; WizardSizePercent=100,80
+
+#if GetEnv("KEY_CONTAINER")
+SignTool=MySignTool
+SignedUninstaller=yes
+#endif
+
+SetupMutex=OllamaSetupMutex
+
+[Languages]
+Name: "english"; MessagesFile: "compiler:Default.isl"
+
+[LangOptions]
+DialogFontSize=12
+
+[Files]
+Source: ".\app.exe"; DestDir: "{app}"; DestName: "{#MyAppExeName}" ; Flags: ignoreversion 64bit
+Source: "..\ollama.exe"; DestDir: "{app}"; Flags: ignoreversion 64bit
+Source: "..\dist\windows-amd64\*.dll"; DestDir: "{app}"; Flags: ignoreversion 64bit
+Source: "..\dist\windows-amd64\ollama_runners\*"; DestDir: "{app}\ollama_runners"; Flags: ignoreversion 64bit recursesubdirs
+Source: "..\dist\ollama_welcome.ps1"; DestDir: "{app}"; Flags: ignoreversion
+Source: ".\assets\app.ico"; DestDir: "{app}"; Flags: ignoreversion
+#if DirExists("..\dist\windows-amd64\rocm")
+  Source: "..\dist\windows-amd64\rocm\*"; DestDir: "{app}\rocm\"; Flags: ignoreversion recursesubdirs
+#endif
+
+
+[Icons]
+Name: "{group}\{#MyAppName}"; Filename: "{app}\{#MyAppExeName}"; IconFilename: "{app}\app.ico"
+Name: "{userstartup}\{#MyAppName}"; Filename: "{app}\{#MyAppExeName}"; IconFilename: "{app}\app.ico"
+Name: "{userprograms}\{#MyAppName}"; Filename: "{app}\{#MyAppExeName}"; IconFilename: "{app}\app.ico"
+
+[Run]
+Filename: "{cmd}"; Parameters: "/C set PATH={app};%PATH% & ""{app}\{#MyAppExeName}"""; Flags: postinstall nowait runhidden
+
+[UninstallRun]
+; Filename: "{cmd}"; Parameters: "/C ""taskkill /im ''{#MyAppExeName}'' /f /t"; Flags: runhidden
+; Filename: "{cmd}"; Parameters: "/C ""taskkill /im ollama.exe /f /t"; Flags: runhidden
+Filename: "taskkill"; Parameters: "/im ""{#MyAppExeName}"" /f /t"; Flags: runhidden
+Filename: "taskkill"; Parameters: "/im ""ollama.exe"" /f /t"; Flags: runhidden
+; HACK!  need to give the server and app enough time to exit
+; TODO - convert this to a Pascal code script so it waits until they're no longer running, then completes
+Filename: "{cmd}"; Parameters: "/c timeout 5"; Flags: runhidden
+
+[UninstallDelete]
+Type: filesandordirs; Name: "{%TEMP}\ollama*"
+Type: filesandordirs; Name: "{%LOCALAPPDATA}\Ollama"
+Type: filesandordirs; Name: "{%LOCALAPPDATA}\Programs\Ollama"
+Type: filesandordirs; Name: "{%USERPROFILE}\.ollama\models"
+Type: filesandordirs; Name: "{%USERPROFILE}\.ollama\history"
+; NOTE: if the user has a custom OLLAMA_MODELS it will be preserved
+
+[Messages]
+WizardReady=Ollama Windows Preview
+ReadyLabel1=%nLet's get you up and running with your own large language models.
+SetupAppRunningError=Another Ollama installer is running.%n%nPlease cancel or finish the other installer, then click OK to continue with this install, or Cancel to exit.
+
+
+;FinishedHeadingLabel=Run your first model
+;FinishedLabel=%nRun this command in a PowerShell or cmd terminal.%n%n%n    ollama run llama3
+;ClickFinish=%n
+
+[Registry]
+Root: HKCU; Subkey: "Environment"; \
+    ValueType: expandsz; ValueName: "Path"; ValueData: "{olddata};{app}"; \
+    Check: NeedsAddPath('{app}')
+
+[Code]
+
+function NeedsAddPath(Param: string): boolean;
+var
+  OrigPath: string;
+begin
+  if not RegQueryStringValue(HKEY_CURRENT_USER,
+    'Environment',
+    'Path', OrigPath)
+  then begin
+    Result := True;
+    exit;
+  end;
+  { look for the path with leading and trailing semicolon }
+  { Pos() returns 0 if not found }
+  Result := Pos(';' + ExpandConstant(Param) + ';', ';' + OrigPath + ';') = 0;
+end;

app/windows/ollama.rc

@@ -0,0 +1,29 @@
+#include <winver.h>
+
+VS_VERSION_INFO VERSIONINFO
+ FILEFLAGSMASK 0x3fL
+#ifdef _DEBUG
+ FILEFLAGS 0x1L
+#else
+ FILEFLAGS 0x0L
+#endif
+ FILEOS 0x40004L
+ FILETYPE 0x1L
+ FILESUBTYPE 0x0L
+BEGIN
+    BLOCK "StringFileInfo"
+    BEGIN
+        BLOCK "040904b0"
+        BEGIN
+            VALUE "FileDescription", "Ollama"
+            VALUE "InternalName", "Ollama"
+            VALUE "OriginalFilename", "ollama app.exe"
+            VALUE "ProductName", "Ollama"
+        END
+    END
+
+    BLOCK "VarFileInfo"
+    BEGIN
+        VALUE "Translation", 0x409, 1200
+    END
+END

app/windows/ollama_welcome.ps1

@@ -0,0 +1,8 @@
+# TODO - consider ANSI colors and maybe ASCII art...
+write-host ""
+write-host "Welcome to Ollama!"
+write-host ""
+write-host "Run your first model:"
+write-host ""
+write-host "`tollama run llama2"
+write-host ""

auth/auth.go

@@ -0,0 +1,61 @@
+package auth
+
+import (
+	"bytes"
+	"context"
+	"crypto/rand"
+	"encoding/base64"
+	"fmt"
+	"io"
+	"log/slog"
+	"os"
+	"path/filepath"
+
+	"golang.org/x/crypto/ssh"
+)
+
+const defaultPrivateKey = "id_ed25519"
+
+func NewNonce(r io.Reader, length int) (string, error) {
+	nonce := make([]byte, length)
+	if _, err := io.ReadFull(r, nonce); err != nil {
+		return "", err
+	}
+
+	return base64.RawURLEncoding.EncodeToString(nonce), nil
+}
+
+func Sign(ctx context.Context, bts []byte) (string, error) {
+	home, err := os.UserHomeDir()
+	if err != nil {
+		return "", err
+	}
+
+	keyPath := filepath.Join(home, ".ollama", defaultPrivateKey)
+
+	privateKeyFile, err := os.ReadFile(keyPath)
+	if err != nil {
+		slog.Info(fmt.Sprintf("Failed to load private key: %v", err))
+		return "", err
+	}
+
+	privateKey, err := ssh.ParsePrivateKey(privateKeyFile)
+	if err != nil {
+		return "", err
+	}
+
+	// get the pubkey, but remove the type
+	publicKey := ssh.MarshalAuthorizedKey(privateKey.PublicKey())
+	parts := bytes.Split(publicKey, []byte(" "))
+	if len(parts) < 2 {
+		return "", fmt.Errorf("malformed public key")
+	}
+
+	signedData, err := privateKey.Sign(rand.Reader, bts)
+	if err != nil {
+		return "", err
+	}
+
+	// signature is <pubkey>:<signature>
+	return fmt.Sprintf("%s:%s", bytes.TrimSpace(parts[1]), base64.StdEncoding.EncodeToString(signedData.Blob)), nil
+}

cmd/cmd.go

@@ -1,6 +1,7 @@
 package cmd
 
 import (
+	"archive/zip"
 	"bytes"
 	"context"
 	"crypto/ed25519"
@@ -14,25 +15,28 @@ import (
 	"net"
 	"net/http"
 	"os"
-	"os/exec"
 	"os/signal"
 	"path/filepath"
+	"regexp"
 	"runtime"
 	"strings"
 	"syscall"
 	"time"
 
+	"github.com/containerd/console"
+
 	"github.com/olekukonko/tablewriter"
 	"github.com/spf13/cobra"
 	"golang.org/x/crypto/ssh"
+	"golang.org/x/exp/slices"
 	"golang.org/x/term"
 
-	"github.com/jmorganca/ollama/api"
+	"github.com/ollama/ollama/api"
-	"github.com/jmorganca/ollama/format"
+	"github.com/ollama/ollama/format"
-	"github.com/jmorganca/ollama/parser"
+	"github.com/ollama/ollama/parser"
-	"github.com/jmorganca/ollama/progress"
+	"github.com/ollama/ollama/progress"
-	"github.com/jmorganca/ollama/server"
+	"github.com/ollama/ollama/server"
-	"github.com/jmorganca/ollama/version"
+	"github.com/ollama/ollama/version"
 )
 
 func CreateHandler(cmd *cobra.Command, args []string) error {
@@ -50,8 +54,6 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 	p := progress.NewProgress(os.Stderr)
 	defer p.Stop()
 
-	bars := make(map[string]*progress.Bar)
-
 	modelfile, err := os.ReadFile(filename)
 	if err != nil {
 		return err
@@ -85,29 +87,41 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 				path = filepath.Join(filepath.Dir(filename), path)
 			}
 
-			bin, err := os.Open(path)
+			fi, err := os.Stat(path)
 			if errors.Is(err, os.ErrNotExist) && c.Name == "model" {
 				continue
 			} else if err != nil {
 				return err
 			}
-			defer bin.Close()
 
-			hash := sha256.New()
+			if fi.IsDir() {
-			if _, err := io.Copy(hash, bin); err != nil {
+				// this is likely a safetensors or pytorch directory
-				return err
+				// TODO make this work w/ adapters
+				tempfile, err := tempZipFiles(path)
+				if err != nil {
+					return err
+				}
+				defer os.RemoveAll(tempfile)
+
+				path = tempfile
 			}
-			bin.Seek(0, io.SeekStart)
 
-			digest := fmt.Sprintf("sha256:%x", hash.Sum(nil))
+			digest, err := createBlob(cmd, client, path)
-			if err = client.CreateBlob(cmd.Context(), digest, bin); err != nil {
+			if err != nil {
 				return err
 			}
 
-			modelfile = bytes.ReplaceAll(modelfile, []byte(c.Args), []byte("@"+digest))
+			name := c.Name
+			if c.Name == "model" {
+				name = "from"
+			}
+
+			re := regexp.MustCompile(fmt.Sprintf(`(?im)^(%s)\s+%s\s*$`, name, c.Args))
+			modelfile = re.ReplaceAll(modelfile, []byte("$1 @"+digest))
 		}
 	}
 
+	bars := make(map[string]*progress.Bar)
 	fn := func(resp api.ProgressResponse) error {
 		if resp.Digest != "" {
 			spinner.Stop()
@@ -131,7 +145,9 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 		return nil
 	}
 
-	request := api.CreateRequest{Name: args[0], Modelfile: string(modelfile)}
+	quantization, _ := cmd.Flags().GetString("quantization")
+
+	request := api.CreateRequest{Name: args[0], Modelfile: string(modelfile), Quantization: quantization}
 	if err := client.Create(cmd.Context(), &request, fn); err != nil {
 		return err
 	}
@@ -139,6 +155,137 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 	return nil
 }
 
+func tempZipFiles(path string) (string, error) {
+	tempfile, err := os.CreateTemp("", "ollama-tf")
+	if err != nil {
+		return "", err
+	}
+	defer tempfile.Close()
+
+	zipfile := zip.NewWriter(tempfile)
+	defer zipfile.Close()
+
+	detectContentType := func(path string) (string, error) {
+		f, err := os.Open(path)
+		if err != nil {
+			return "", err
+		}
+		defer f.Close()
+
+		var b bytes.Buffer
+		b.Grow(512)
+
+		if _, err := io.CopyN(&b, f, 512); err != nil && !errors.Is(err, io.EOF) {
+			return "", err
+		}
+
+		contentType, _, _ := strings.Cut(http.DetectContentType(b.Bytes()), ";")
+		return contentType, nil
+	}
+
+	glob := func(pattern, contentType string) ([]string, error) {
+		matches, err := filepath.Glob(pattern)
+		if err != nil {
+			return nil, err
+		}
+
+		for _, safetensor := range matches {
+			if ct, err := detectContentType(safetensor); err != nil {
+				return nil, err
+			} else if ct != contentType {
+				return nil, fmt.Errorf("invalid content type: expected %s for %s", ct, safetensor)
+			}
+		}
+
+		return matches, nil
+	}
+
+	var files []string
+	if st, _ := glob(filepath.Join(path, "model*.safetensors"), "application/octet-stream"); len(st) > 0 {
+		// safetensors files might be unresolved git lfs references; skip if they are
+		// covers model-x-of-y.safetensors, model.fp32-x-of-y.safetensors, model.safetensors
+		files = append(files, st...)
+	} else if pt, _ := glob(filepath.Join(path, "pytorch_model*.bin"), "application/zip"); len(pt) > 0 {
+		// pytorch files might also be unresolved git lfs references; skip if they are
+		// covers pytorch_model-x-of-y.bin, pytorch_model.fp32-x-of-y.bin, pytorch_model.bin
+		files = append(files, pt...)
+	} else if pt, _ := glob(filepath.Join(path, "consolidated*.pth"), "application/octet-stream"); len(pt) > 0 {
+		// pytorch files might also be unresolved git lfs references; skip if they are
+		// covers consolidated.x.pth, consolidated.pth
+		files = append(files, pt...)
+	} else {
+		return "", errors.New("no safetensors or torch files found")
+	}
+
+	// add configuration files, json files are detected as text/plain
+	js, err := glob(filepath.Join(path, "*.json"), "text/plain")
+	if err != nil {
+		return "", err
+	}
+	files = append(files, js...)
+
+	if tks, _ := glob(filepath.Join(path, "tokenizer.model"), "application/octet-stream"); len(tks) > 0 {
+		// add tokenizer.model if it exists, tokenizer.json is automatically picked up by the previous glob
+		// tokenizer.model might be a unresolved git lfs reference; error if it is
+		files = append(files, tks...)
+	} else if tks, _ := glob(filepath.Join(path, "**/tokenizer.model"), "text/plain"); len(tks) > 0 {
+		// some times tokenizer.model is in a subdirectory (e.g. meta-llama/Meta-Llama-3-8B)
+		files = append(files, tks...)
+	}
+
+	for _, file := range files {
+		f, err := os.Open(file)
+		if err != nil {
+			return "", err
+		}
+		defer f.Close()
+
+		fi, err := f.Stat()
+		if err != nil {
+			return "", err
+		}
+
+		zfi, err := zip.FileInfoHeader(fi)
+		if err != nil {
+			return "", err
+		}
+
+		zf, err := zipfile.CreateHeader(zfi)
+		if err != nil {
+			return "", err
+		}
+
+		if _, err := io.Copy(zf, f); err != nil {
+			return "", err
+		}
+	}
+
+	return tempfile.Name(), nil
+}
+
+func createBlob(cmd *cobra.Command, client *api.Client, path string) (string, error) {
+	bin, err := os.Open(path)
+	if err != nil {
+		return "", err
+	}
+	defer bin.Close()
+
+	hash := sha256.New()
+	if _, err := io.Copy(hash, bin); err != nil {
+		return "", err
+	}
+
+	if _, err := bin.Seek(0, io.SeekStart); err != nil {
+		return "", err
+	}
+
+	digest := fmt.Sprintf("sha256:%x", hash.Sum(nil))
+	if err = client.CreateBlob(cmd.Context(), digest, bin); err != nil {
+		return "", err
+	}
+	return digest, nil
+}
+
 func RunHandler(cmd *cobra.Command, args []string) error {
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
@@ -146,19 +293,68 @@ func RunHandler(cmd *cobra.Command, args []string) error {
 	}
 
 	name := args[0]
+
 	// check if the model exists on the server
-	_, err = client.Show(cmd.Context(), &api.ShowRequest{Name: name})
+	show, err := client.Show(cmd.Context(), &api.ShowRequest{Name: name})
 	var statusError api.StatusError
 	switch {
 	case errors.As(err, &statusError) && statusError.StatusCode == http.StatusNotFound:
 		if err := PullHandler(cmd, []string{name}); err != nil {
 			return err
 		}
+
+		show, err = client.Show(cmd.Context(), &api.ShowRequest{Name: name})
+		if err != nil {
+			return err
+		}
 	case err != nil:
 		return err
 	}
 
-	return RunGenerate(cmd, args)
+	interactive := true
+
+	opts := runOptions{
+		Model:       args[0],
+		WordWrap:    os.Getenv("TERM") == "xterm-256color",
+		Options:     map[string]interface{}{},
+		MultiModal:  slices.Contains(show.Details.Families, "clip"),
+		ParentModel: show.Details.ParentModel,
+	}
+
+	format, err := cmd.Flags().GetString("format")
+	if err != nil {
+		return err
+	}
+	opts.Format = format
+
+	prompts := args[1:]
+	// prepend stdin to the prompt if provided
+	if !term.IsTerminal(int(os.Stdin.Fd())) {
+		in, err := io.ReadAll(os.Stdin)
+		if err != nil {
+			return err
+		}
+
+		prompts = append([]string{string(in)}, prompts...)
+		opts.WordWrap = false
+		interactive = false
+	}
+	opts.Prompt = strings.Join(prompts, " ")
+	if len(prompts) > 0 {
+		interactive = false
+	}
+
+	nowrap, err := cmd.Flags().GetBool("nowordwrap")
+	if err != nil {
+		return err
+	}
+	opts.WordWrap = !nowrap
+
+	if !interactive {
+		return generate(cmd, opts)
+	}
+
+	return generateInteractive(cmd, opts)
 }
 
 func PushHandler(cmd *cobra.Command, args []string) error {
@@ -410,63 +606,20 @@ func PullHandler(cmd *cobra.Command, args []string) error {
 	return nil
 }
 
-func RunGenerate(cmd *cobra.Command, args []string) error {
-	interactive := true
-
-	opts := runOptions{
-		Model:    args[0],
-		WordWrap: os.Getenv("TERM") == "xterm-256color",
-		Options:  map[string]interface{}{},
-	}
-
-	format, err := cmd.Flags().GetString("format")
-	if err != nil {
-		return err
-	}
-	opts.Format = format
-
-	prompts := args[1:]
-	// prepend stdin to the prompt if provided
-	if !term.IsTerminal(int(os.Stdin.Fd())) {
-		in, err := io.ReadAll(os.Stdin)
-		if err != nil {
-			return err
-		}
-
-		prompts = append([]string{string(in)}, prompts...)
-		opts.WordWrap = false
-		interactive = false
-	}
-	opts.Prompt = strings.Join(prompts, " ")
-	if len(prompts) > 0 {
-		interactive = false
-	}
-
-	nowrap, err := cmd.Flags().GetBool("nowordwrap")
-	if err != nil {
-		return err
-	}
-	opts.WordWrap = !nowrap
-
-	if !interactive {
-		return generate(cmd, opts)
-	}
-
-	return generateInteractive(cmd, opts)
-}
-
 type generateContextKey string
 
 type runOptions struct {
-	Model    string
+	Model       string
-	Prompt   string
+	ParentModel string
-	Messages []api.Message
+	Prompt      string
-	WordWrap bool
+	Messages    []api.Message
-	Format   string
+	WordWrap    bool
-	System   string
+	Format      string
-	Template string
+	System      string
-	Images   []api.ImageData
+	Template    string
-	Options  map[string]interface{}
+	Images      []api.ImageData
+	Options     map[string]interface{}
+	MultiModal  bool
 }
 
 type displayResponseState struct {
@@ -628,10 +781,18 @@ func generate(cmd *cobra.Command, opts runOptions) error {
 		return nil
 	}
 
+	if opts.MultiModal {
+		opts.Prompt, opts.Images, err = extractFileData(opts.Prompt)
+		if err != nil {
+			return err
+		}
+	}
+
 	request := api.GenerateRequest{
 		Model:    opts.Model,
 		Prompt:   opts.Prompt,
 		Context:  generateContext,
+		Images:   opts.Images,
 		Format:   opts.Format,
 		System:   opts.System,
 		Template: opts.Template,
@@ -670,7 +831,7 @@ func generate(cmd *cobra.Command, opts runOptions) error {
 }
 
 func RunServer(cmd *cobra.Command, _ []string) error {
-	host, port, err := net.SplitHostPort(os.Getenv("OLLAMA_HOST"))
+	host, port, err := net.SplitHostPort(strings.Trim(os.Getenv("OLLAMA_HOST"), "\"'"))
 	if err != nil {
 		host, port = "127.0.0.1", "11434"
 		if ip := net.ParseIP(strings.Trim(os.Getenv("OLLAMA_HOST"), "[]")); ip != nil {
@@ -702,59 +863,42 @@ func initializeKeypair() error {
 	_, err = os.Stat(privKeyPath)
 	if os.IsNotExist(err) {
 		fmt.Printf("Couldn't find '%s'. Generating new private key.\n", privKeyPath)
-		_, privKey, err := ed25519.GenerateKey(rand.Reader)
+		cryptoPublicKey, cryptoPrivateKey, err := ed25519.GenerateKey(rand.Reader)
 		if err != nil {
 			return err
 		}
 
-		privKeyBytes, err := format.OpenSSHPrivateKey(privKey, "")
+		privateKeyBytes, err := ssh.MarshalPrivateKey(cryptoPrivateKey, "")
 		if err != nil {
 			return err
 		}
 
-		err = os.MkdirAll(filepath.Dir(privKeyPath), 0o755)
+		if err := os.MkdirAll(filepath.Dir(privKeyPath), 0o755); err != nil {
-		if err != nil {
 			return fmt.Errorf("could not create directory %w", err)
 		}
 
-		err = os.WriteFile(privKeyPath, pem.EncodeToMemory(privKeyBytes), 0o600)
+		if err := os.WriteFile(privKeyPath, pem.EncodeToMemory(privateKeyBytes), 0o600); err != nil {
+			return err
+		}
+
+		sshPublicKey, err := ssh.NewPublicKey(cryptoPublicKey)
 		if err != nil {
 			return err
 		}
 
-		sshPrivateKey, err := ssh.NewSignerFromKey(privKey)
+		publicKeyBytes := ssh.MarshalAuthorizedKey(sshPublicKey)
-		if err != nil {
+
+		if err := os.WriteFile(pubKeyPath, publicKeyBytes, 0o644); err != nil {
 			return err
 		}
 
-		pubKeyData := ssh.MarshalAuthorizedKey(sshPrivateKey.PublicKey())
+		fmt.Printf("Your new public key is: \n\n%s\n", publicKeyBytes)
-
-		err = os.WriteFile(pubKeyPath, pubKeyData, 0o644)
-		if err != nil {
-			return err
-		}
-
-		fmt.Printf("Your new public key is: \n\n%s\n", string(pubKeyData))
 	}
 	return nil
 }
 
-func startMacApp(ctx context.Context, client *api.Client) error {
+//nolint:unused
-	exe, err := os.Executable()
+func waitForServer(ctx context.Context, client *api.Client) error {
-	if err != nil {
-		return err
-	}
-	link, err := os.Readlink(exe)
-	if err != nil {
-		return err
-	}
-	if !strings.Contains(link, "Ollama.app") {
-		return fmt.Errorf("could not find ollama app")
-	}
-	path := strings.Split(link, "Ollama.app")
-	if err := exec.Command("/usr/bin/open", "-a", path[0]+"Ollama.app").Run(); err != nil {
-		return err
-	}
 	// wait for the server to start
 	timeout := time.After(5 * time.Second)
 	tick := time.Tick(500 * time.Millisecond)
@@ -768,6 +912,7 @@ func startMacApp(ctx context.Context, client *api.Client) error {
 			}
 		}
 	}
+
 }
 
 func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
@@ -776,15 +921,11 @@ func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
 		return err
 	}
 	if err := client.Heartbeat(cmd.Context()); err != nil {
-		if !strings.Contains(err.Error(), "connection refused") {
+		if !strings.Contains(err.Error(), " refused") {
 			return err
 		}
-		if runtime.GOOS == "darwin" {
+		if err := startApp(cmd.Context(), client); err != nil {
-			if err := startMacApp(cmd.Context(), client); err != nil {
+			return fmt.Errorf("could not connect to ollama app, is it running?")
-				return fmt.Errorf("could not connect to ollama app, is it running?")
-			}
-		} else {
-			return fmt.Errorf("could not connect to ollama server, run 'ollama serve' to start it")
 		}
 	}
 	return nil
@@ -810,10 +951,22 @@ func versionHandler(cmd *cobra.Command, _ []string) {
 	}
 }
 
+func appendHostEnvDocs(cmd *cobra.Command) {
+	const hostEnvDocs = `
+Environment Variables:
+      OLLAMA_HOST        The host:port or base URL of the Ollama server (e.g. http://localhost:11434)
+`
+	cmd.SetUsageTemplate(cmd.UsageTemplate() + hostEnvDocs)
+}
+
 func NewCLI() *cobra.Command {
 	log.SetFlags(log.LstdFlags | log.Lshortfile)
 	cobra.EnableCommandSorting = false
 
+	if runtime.GOOS == "windows" {
+		console.ConsoleFromFile(os.Stdin) //nolint:errcheck
+	}
+
 	rootCmd := &cobra.Command{
 		Use:           "ollama",
 		Short:         "Large language model runner",
@@ -843,6 +996,7 @@ func NewCLI() *cobra.Command {
 	}
 
 	createCmd.Flags().StringP("file", "f", "Modelfile", "Name of the Modelfile (default \"Modelfile\")")
+	createCmd.Flags().StringP("quantization", "q", "", "Quantization level.")
 
 	showCmd := &cobra.Command{
 		Use:     "show MODEL",
@@ -870,7 +1024,6 @@ func NewCLI() *cobra.Command {
 	runCmd.Flags().Bool("insecure", false, "Use an insecure registry")
 	runCmd.Flags().Bool("nowordwrap", false, "Don't wrap words to the next line automatically")
 	runCmd.Flags().String("format", "", "Response format (e.g. json)")
-
 	serveCmd := &cobra.Command{
 		Use:     "serve",
 		Aliases: []string{"start"},
@@ -878,6 +1031,15 @@ func NewCLI() *cobra.Command {
 		Args:    cobra.ExactArgs(0),
 		RunE:    RunServer,
 	}
+	serveCmd.SetUsageTemplate(serveCmd.UsageTemplate() + `
+Environment Variables:
+
+    OLLAMA_HOST         The host:port to bind to (default "127.0.0.1:11434")
+    OLLAMA_ORIGINS      A comma separated list of allowed origins.
+    OLLAMA_MODELS       The path to the models directory (default is "~/.ollama/models")
+    OLLAMA_KEEP_ALIVE   The duration that models stay loaded in memory (default is "5m")
+    OLLAMA_DEBUG        Set to 1 to enable additional debug logging
+`)
 
 	pullCmd := &cobra.Command{
 		Use:     "pull MODEL",
@@ -906,7 +1068,6 @@ func NewCLI() *cobra.Command {
 		PreRunE: checkServerHeartbeat,
 		RunE:    ListHandler,
 	}
-
 	copyCmd := &cobra.Command{
 		Use:     "cp SOURCE TARGET",
 		Short:   "Copy a model",
@@ -923,6 +1084,19 @@ func NewCLI() *cobra.Command {
 		RunE:    DeleteHandler,
 	}
 
+	for _, cmd := range []*cobra.Command{
+		createCmd,
+		showCmd,
+		runCmd,
+		pullCmd,
+		pushCmd,
+		listCmd,
+		copyCmd,
+		deleteCmd,
+	} {
+		appendHostEnvDocs(cmd)
+	}
+
 	rootCmd.AddCommand(
 		serveCmd,
 		createCmd,

cmd/interactive.go

@@ -6,14 +6,17 @@ import (
 	"io"
 	"net/http"
 	"os"
+	"path/filepath"
 	"regexp"
+	"sort"
 	"strings"
 
 	"github.com/spf13/cobra"
 	"golang.org/x/exp/slices"
 
-	"github.com/jmorganca/ollama/api"
+	"github.com/ollama/ollama/api"
-	"github.com/jmorganca/ollama/readline"
+	"github.com/ollama/ollama/progress"
+	"github.com/ollama/ollama/readline"
 )
 
 type MultilineState int
@@ -25,45 +28,82 @@ const (
 	MultilineTemplate
 )
 
-func modelIsMultiModal(cmd *cobra.Command, name string) bool {
+func loadModel(cmd *cobra.Command, opts *runOptions) error {
-	// get model details
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
-		fmt.Println("error: couldn't connect to ollama server")
+		return err
-		return false
 	}
 
-	req := api.ShowRequest{Name: name}
+	p := progress.NewProgress(os.Stderr)
-	resp, err := client.Show(cmd.Context(), &req)
+	defer p.StopAndClear()
+
+	spinner := progress.NewSpinner("")
+	p.Add("", spinner)
+
+	showReq := api.ShowRequest{Name: opts.Model}
+	showResp, err := client.Show(cmd.Context(), &showReq)
 	if err != nil {
-		return false
+		return err
+	}
+	opts.MultiModal = slices.Contains(showResp.Details.Families, "clip")
+	opts.ParentModel = showResp.Details.ParentModel
+
+	if len(showResp.Messages) > 0 {
+		opts.Messages = append(opts.Messages, showResp.Messages...)
 	}
 
-	return slices.Contains(resp.Details.Families, "clip")
+	chatReq := &api.ChatRequest{
+		Model:    opts.Model,
+		Messages: []api.Message{},
+	}
+	err = client.Chat(cmd.Context(), chatReq, func(resp api.ChatResponse) error {
+		p.StopAndClear()
+		if len(opts.Messages) > 0 {
+			for _, msg := range opts.Messages {
+				switch msg.Role {
+				case "user":
+					fmt.Printf(">>> %s\n", msg.Content)
+				case "assistant":
+					state := &displayResponseState{}
+					displayResponse(msg.Content, opts.WordWrap, state)
+					fmt.Println()
+					fmt.Println()
+				}
+			}
+		}
+		return nil
+	})
+	if err != nil {
+		return err
+	}
+
+	return nil
 }
 
 func generateInteractive(cmd *cobra.Command, opts runOptions) error {
-	multiModal := modelIsMultiModal(cmd, opts.Model)
+	opts.Messages = make([]api.Message, 0)
 
-	// load the model
+	err := loadModel(cmd, &opts)
-	loadOpts := runOptions{
+	if err != nil {
-		Model:    opts.Model,
-		Prompt:   "",
-		Messages: []api.Message{},
-	}
-	if _, err := chat(cmd, loadOpts); err != nil {
 		return err
 	}
 
 	usage := func() {
 		fmt.Fprintln(os.Stderr, "Available Commands:")
-		fmt.Fprintln(os.Stderr, "  /set          Set session variables")
+		fmt.Fprintln(os.Stderr, "  /set            Set session variables")
-		fmt.Fprintln(os.Stderr, "  /show         Show model information")
+		fmt.Fprintln(os.Stderr, "  /show           Show model information")
-		fmt.Fprintln(os.Stderr, "  /bye          Exit")
+		fmt.Fprintln(os.Stderr, "  /load <model>   Load a session or model")
-		fmt.Fprintln(os.Stderr, "  /?, /help     Help for a command")
+		fmt.Fprintln(os.Stderr, "  /save <model>   Save your current session")
-		fmt.Fprintln(os.Stderr, "  /? shortcuts  Help for keyboard shortcuts")
+		fmt.Fprintln(os.Stderr, "  /bye            Exit")
+		fmt.Fprintln(os.Stderr, "  /?, /help       Help for a command")
+		fmt.Fprintln(os.Stderr, "  /? shortcuts    Help for keyboard shortcuts")
 		fmt.Fprintln(os.Stderr, "")
 		fmt.Fprintln(os.Stderr, "Use \"\"\" to begin a multi-line message.")
+
+		if opts.MultiModal {
+			fmt.Fprintf(os.Stderr, "Use %s to include .jpg or .png images.\n", filepath.FromSlash("/path/to/file"))
+		}
+
 		fmt.Fprintln(os.Stderr, "")
 	}
 
@@ -140,7 +180,6 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 
 	var sb strings.Builder
 	var multiline MultilineState
-	opts.Messages = make([]api.Message, 0)
 
 	for {
 		line, err := scanner.Readline()
@@ -174,6 +213,7 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 			switch multiline {
 			case MultilineSystem:
 				opts.System = sb.String()
+				opts.Messages = append(opts.Messages, api.Message{Role: "system", Content: opts.System})
 				fmt.Println("Set system message.")
 				sb.Reset()
 			case MultilineTemplate:
@@ -193,7 +233,6 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 				fmt.Fprintln(&sb)
 				multiline = MultilinePrompt
 				scanner.Prompt.UseAlt = true
-				break
 			}
 		case scanner.Pasting:
 			fmt.Fprintln(&sb, line)
@@ -203,6 +242,44 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 			if err := ListHandler(cmd, args[1:]); err != nil {
 				return err
 			}
+		case strings.HasPrefix(line, "/load"):
+			args := strings.Fields(line)
+			if len(args) != 2 {
+				fmt.Println("Usage:\n  /load <modelname>")
+				continue
+			}
+			opts.Model = args[1]
+			opts.Messages = []api.Message{}
+			fmt.Printf("Loading model '%s'\n", opts.Model)
+			if err := loadModel(cmd, &opts); err != nil {
+				return err
+			}
+			continue
+		case strings.HasPrefix(line, "/save"):
+			args := strings.Fields(line)
+			if len(args) != 2 {
+				fmt.Println("Usage:\n  /save <modelname>")
+				continue
+			}
+
+			client, err := api.ClientFromEnvironment()
+			if err != nil {
+				fmt.Println("error: couldn't connect to ollama server")
+				return err
+			}
+
+			req := &api.CreateRequest{
+				Name:      args[1],
+				Modelfile: buildModelfile(opts),
+			}
+			fn := func(resp api.ProgressResponse) error { return nil }
+			err = client.Create(cmd.Context(), req, fn)
+			if err != nil {
+				fmt.Println("error: couldn't save model")
+				return err
+			}
+			fmt.Printf("Created new model '%s'\n", args[1])
+			continue
 		case strings.HasPrefix(line, "/set"):
 			args := strings.Fields(line)
 			if len(args) > 1 {
@@ -218,10 +295,14 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 					opts.WordWrap = false
 					fmt.Println("Set 'nowordwrap' mode.")
 				case "verbose":
-					cmd.Flags().Set("verbose", "true")
+					if err := cmd.Flags().Set("verbose", "true"); err != nil {
+						return err
+					}
 					fmt.Println("Set 'verbose' mode.")
 				case "quiet":
-					cmd.Flags().Set("verbose", "false")
+					if err := cmd.Flags().Set("verbose", "false"); err != nil {
+						return err
+					}
 					fmt.Println("Set 'quiet' mode.")
 				case "format":
 					if len(args) < 3 || args[2] != "json" {
@@ -277,11 +358,21 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 					}
 
 					if args[1] == "system" {
-						opts.System = sb.String()
+						opts.System = sb.String() // for display in modelfile
+						newMessage := api.Message{Role: "system", Content: sb.String()}
+						// Check if the slice is not empty and the last message is from 'system'
+						if len(opts.Messages) > 0 && opts.Messages[len(opts.Messages)-1].Role == "system" {
+							// Replace the last message
+							opts.Messages[len(opts.Messages)-1] = newMessage
+						} else {
+							opts.Messages = append(opts.Messages, newMessage)
+						}
 						fmt.Println("Set system message.")
+						sb.Reset()
 					} else if args[1] == "template" {
 						opts.Template = sb.String()
 						fmt.Println("Set prompt template.")
+						sb.Reset()
 					}
 
 					sb.Reset()
@@ -383,13 +474,13 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 			} else {
 				usage()
 			}
-		case line == "/exit", line == "/bye":
+		case strings.HasPrefix(line, "/exit"), strings.HasPrefix(line, "/bye"):
 			return nil
 		case strings.HasPrefix(line, "/"):
 			args := strings.Fields(line)
 			isFile := false
 
-			if multiModal {
+			if opts.MultiModal {
 				for _, f := range extractFileNames(line) {
 					if strings.HasPrefix(f, args[0]) {
 						isFile = true
@@ -411,34 +502,23 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 		if sb.Len() > 0 && multiline == MultilineNone {
 			newMessage := api.Message{Role: "user", Content: sb.String()}
 
-			if multiModal {
+			if opts.MultiModal {
 				msg, images, err := extractFileData(sb.String())
 				if err != nil {
 					return err
 				}
-				newMessage.Content = msg
 
-				// reset the context if we find another image
+				// clear all previous images for better responses
 				if len(images) > 0 {
-					newMessage.Images = append(newMessage.Images, images...)
+					for i := range opts.Messages {
-					// reset the context for the new image
+						opts.Messages[i].Images = nil
-					opts.Messages = []api.Message{}
-				} else {
-					if len(opts.Messages) > 1 {
-						newMessage.Images = append(newMessage.Images, opts.Messages[len(opts.Messages)-2].Images...)
 					}
 				}
-				if len(newMessage.Images) == 0 {
+
-					fmt.Println("This model requires you to add a jpeg, png, or svg image.")
+				newMessage.Content = msg
-					fmt.Println()
+				newMessage.Images = images
-					sb.Reset()
-					continue
-				}
 			}
 
-			if opts.System != "" {
-				opts.Messages = append(opts.Messages, api.Message{Role: "system", Content: opts.System})
-			}
 			opts.Messages = append(opts.Messages, newMessage)
 
 			assistant, err := chat(cmd, opts)
@@ -454,6 +534,38 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 	}
 }
 
+func buildModelfile(opts runOptions) string {
+	var mf strings.Builder
+	model := opts.ParentModel
+	if model == "" {
+		model = opts.Model
+	}
+	fmt.Fprintf(&mf, "FROM %s\n", model)
+	if opts.System != "" {
+		fmt.Fprintf(&mf, "SYSTEM \"\"\"%s\"\"\"\n", opts.System)
+	}
+
+	if opts.Template != "" {
+		fmt.Fprintf(&mf, "TEMPLATE \"\"\"%s\"\"\"\n", opts.Template)
+	}
+
+	keys := make([]string, 0)
+	for k := range opts.Options {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	for _, k := range keys {
+		fmt.Fprintf(&mf, "PARAMETER %s %v\n", k, opts.Options[k])
+	}
+	fmt.Fprintln(&mf)
+
+	for _, msg := range opts.Messages {
+		fmt.Fprintf(&mf, "MESSAGE %s \"\"\"%s\"\"\"\n", msg.Role, msg.Content)
+	}
+
+	return mf.String()
+}
+
 func normalizeFilePath(fp string) string {
 	// Define a map of escaped characters and their replacements
 	replacements := map[string]string{
@@ -500,10 +612,10 @@ func extractFileData(input string) (string, []api.ImageData, error) {
 			if os.IsNotExist(err) {
 				continue
 			}
-			fmt.Printf("Couldn't process image: %q\n", err)
+			fmt.Fprintf(os.Stderr, "Couldn't process image: %q\n", err)
 			return "", imgs, err
 		}
-		fmt.Printf("Added image '%s'\n", nfp)
+		fmt.Fprintf(os.Stderr, "Added image '%s'\n", nfp)
 		input = strings.ReplaceAll(input, fp, "")
 		imgs = append(imgs, data)
 	}
@@ -524,7 +636,7 @@ func getImageData(filePath string) ([]byte, error) {
 	}
 
 	contentType := http.DetectContentType(buf)
-	allowedTypes := []string{"image/jpeg", "image/jpg", "image/svg+xml", "image/png"}
+	allowedTypes := []string{"image/jpeg", "image/jpg", "image/png"}
 	if !slices.Contains(allowedTypes, contentType) {
 		return nil, fmt.Errorf("invalid image type: %s", contentType)
 	}

cmd/interactive_test.go

@@ -1,9 +1,13 @@
 package cmd
 
 import (
+	"bytes"
 	"testing"
+	"text/template"
 
 	"github.com/stretchr/testify/assert"
+
+	"github.com/ollama/ollama/api"
 )
 
 func TestExtractFilenames(t *testing.T) {
@@ -49,3 +53,64 @@ d:\path with\spaces\seven.svg inbetween7 c:\users\jdoe\eight.png inbetween8
 	assert.Contains(t, res[9], "ten.svg")
 	assert.Contains(t, res[9], "E:")
 }
+
+func TestModelfileBuilder(t *testing.T) {
+	opts := runOptions{
+		Model:    "hork",
+		System:   "You are part horse and part shark, but all hork. Do horklike things",
+		Template: "This is a template.",
+		Messages: []api.Message{
+			{Role: "user", Content: "Hey there hork!"},
+			{Role: "assistant", Content: "Yes it is true, I am half horse, half shark."},
+		},
+		Options: map[string]interface{}{},
+	}
+
+	opts.Options["temperature"] = 0.9
+	opts.Options["seed"] = 42
+	opts.Options["penalize_newline"] = false
+	opts.Options["stop"] = []string{"hi", "there"}
+
+	mf := buildModelfile(opts)
+	expectedModelfile := `FROM {{.Model}}
+SYSTEM """{{.System}}"""
+TEMPLATE """{{.Template}}"""
+PARAMETER penalize_newline false
+PARAMETER seed 42
+PARAMETER stop [hi there]
+PARAMETER temperature 0.9
+
+MESSAGE user """Hey there hork!"""
+MESSAGE assistant """Yes it is true, I am half horse, half shark."""
+`
+
+	tmpl, err := template.New("").Parse(expectedModelfile)
+	assert.Nil(t, err)
+
+	var buf bytes.Buffer
+	err = tmpl.Execute(&buf, opts)
+	assert.Nil(t, err)
+	assert.Equal(t, buf.String(), mf)
+
+	opts.ParentModel = "horseshark"
+	mf = buildModelfile(opts)
+	expectedModelfile = `FROM {{.ParentModel}}
+SYSTEM """{{.System}}"""
+TEMPLATE """{{.Template}}"""
+PARAMETER penalize_newline false
+PARAMETER seed 42
+PARAMETER stop [hi there]
+PARAMETER temperature 0.9
+
+MESSAGE user """Hey there hork!"""
+MESSAGE assistant """Yes it is true, I am half horse, half shark."""
+`
+
+	tmpl, err = template.New("").Parse(expectedModelfile)
+	assert.Nil(t, err)
+
+	var parentBuf bytes.Buffer
+	err = tmpl.Execute(&parentBuf, opts)
+	assert.Nil(t, err)
+	assert.Equal(t, parentBuf.String(), mf)
+}

cmd/start_darwin.go

@@ -0,0 +1,30 @@
+package cmd
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"os/exec"
+	"strings"
+
+	"github.com/ollama/ollama/api"
+)
+
+func startApp(ctx context.Context, client *api.Client) error {
+	exe, err := os.Executable()
+	if err != nil {
+		return err
+	}
+	link, err := os.Readlink(exe)
+	if err != nil {
+		return err
+	}
+	if !strings.Contains(link, "Ollama.app") {
+		return fmt.Errorf("could not find ollama app")
+	}
+	path := strings.Split(link, "Ollama.app")
+	if err := exec.Command("/usr/bin/open", "-a", path[0]+"Ollama.app").Run(); err != nil {
+		return err
+	}
+	return waitForServer(ctx, client)
+}

cmd/start_default.go

@@ -0,0 +1,14 @@
+//go:build !windows && !darwin
+
+package cmd
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/ollama/ollama/api"
+)
+
+func startApp(ctx context.Context, client *api.Client) error {
+	return fmt.Errorf("could not connect to ollama server, run 'ollama serve' to start it")
+}

cmd/start_windows.go

@@ -0,0 +1,58 @@
+package cmd
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"syscall"
+
+	"github.com/ollama/ollama/api"
+)
+
+func startApp(ctx context.Context, client *api.Client) error {
+	// log.Printf("XXX Attempting to find and start ollama app")
+	AppName := "ollama app.exe"
+	exe, err := os.Executable()
+	if err != nil {
+		return err
+	}
+	appExe := filepath.Join(filepath.Dir(exe), AppName)
+	_, err = os.Stat(appExe)
+	if errors.Is(err, os.ErrNotExist) {
+		// Try the standard install location
+		localAppData := os.Getenv("LOCALAPPDATA")
+		appExe = filepath.Join(localAppData, "Ollama", AppName)
+		_, err := os.Stat(appExe)
+		if errors.Is(err, os.ErrNotExist) {
+			// Finally look in the path
+			appExe, err = exec.LookPath(AppName)
+			if err != nil {
+				return fmt.Errorf("could not locate ollama app")
+			}
+		}
+	}
+	// log.Printf("XXX attempting to start app %s", appExe)
+
+	cmd_path := "c:\\Windows\\system32\\cmd.exe"
+	cmd := exec.Command(cmd_path, "/c", appExe)
+	// TODO - these hide flags aren't working - still pops up a command window for some reason
+	cmd.SysProcAttr = &syscall.SysProcAttr{CreationFlags: 0x08000000, HideWindow: true}
+
+	// TODO this didn't help either...
+	cmd.Stdin = strings.NewReader("")
+	cmd.Stdout = os.Stdout
+	cmd.Stderr = os.Stderr
+
+	if err := cmd.Start(); err != nil {
+		return fmt.Errorf("unable to start ollama app %w", err)
+	}
+
+	if cmd.Process != nil {
+		defer cmd.Process.Release() //nolint:errcheck
+	}
+	return waitForServer(ctx, client)
+}

convert/convert.go

@@ -0,0 +1,187 @@
+package convert
+
+import (
+	"cmp"
+	"encoding/binary"
+	"encoding/json"
+	"fmt"
+	"log/slog"
+	"os"
+	"path/filepath"
+	"slices"
+	"strings"
+
+	"google.golang.org/protobuf/proto"
+
+	"github.com/ollama/ollama/convert/sentencepiece"
+	"github.com/ollama/ollama/llm"
+)
+
+type Params struct {
+	Architectures     []string `json:"architectures"`
+	VocabSize         int      `json:"vocab_size"`
+	HiddenSize        int      `json:"hidden_size"`       // n_embd
+	HiddenLayers      int      `json:"num_hidden_layers"` // n_layer
+	ContextSize       int      `json:"max_position_embeddings"`
+	IntermediateSize  int      `json:"intermediate_size"`
+	AttentionHeads    int      `json:"num_attention_heads"` // n_head
+	KeyValHeads       int      `json:"num_key_value_heads"`
+	NormEPS           float64  `json:"rms_norm_eps"`
+	BoSTokenID        int      `json:"bos_token_id"`
+	EoSTokenID        int      `json:"eos_token_id"`
+	HeadDimension     int      `json:"head_dim"`
+	PaddingTokenID    int      `json:"pad_token_id"`
+	RopeFrequencyBase float64  `json:"rope_theta"`
+
+	Experts     int `json:"num_local_experts"`
+	ExpertsUsed int `json:"num_experts_per_tok"`
+
+	ByteOrder
+}
+
+type ByteOrder interface {
+	binary.ByteOrder
+	binary.AppendByteOrder
+}
+
+type ModelArch interface {
+	GetTensors() error
+	LoadVocab() error
+	WriteGGUF() (string, error)
+}
+
+type ModelFormat interface {
+	GetLayerName(string) (string, error)
+	GetTensors(string, *Params) ([]llm.Tensor, error)
+	GetParams(string) (*Params, error)
+	GetModelArch(string, string, *Params) (ModelArch, error)
+}
+
+type ModelData struct {
+	Path    string
+	Name    string
+	Params  *Params
+	Vocab   *Vocab
+	Tensors []llm.Tensor
+	Format  ModelFormat
+}
+
+func GetModelFormat(dirname string) (ModelFormat, error) {
+	files, err := filepath.Glob(filepath.Join(dirname, "*"))
+	if err != nil {
+		return nil, err
+	}
+
+	for _, fn := range files {
+		slog.Debug(fmt.Sprintf("file = %s", fn))
+		if strings.HasSuffix(fn, ".safetensors") {
+			return &SafetensorFormat{}, nil
+		} else if strings.HasSuffix(fn, ".bin") {
+			slog.Debug("model is torch")
+			return &TorchFormat{}, nil
+		}
+	}
+
+	return nil, fmt.Errorf("couldn't determine model format")
+}
+
+// Details on gguf's tokenizer can be found at:
+// https://github.com/ggerganov/ggml/blob/master/docs/gguf.md#tokenizer
+type Vocab struct {
+	Tokens []string
+	Scores []float32
+	Types  []int32
+}
+
+func LoadSentencePieceTokens(dirpath string, params *Params) (*Vocab, error) {
+	slog.Info(fmt.Sprintf("reading vocab from %s", filepath.Join(dirpath, "tokenizer.model")))
+	in, err := os.ReadFile(filepath.Join(dirpath, "tokenizer.model"))
+	if err != nil {
+		return nil, err
+	}
+
+	// To regenerate sentencepiece from the protobufs use:
+	// protoc -I=./ --go_out=./ sentencepiece_model.proto
+	modelProto := &sentencepiece.ModelProto{}
+	if err := proto.Unmarshal(in, modelProto); err != nil {
+		return nil, err
+	}
+
+	v := &Vocab{
+		Tokens: make([]string, 0),
+		Scores: make([]float32, 0),
+		Types:  make([]int32, 0),
+	}
+
+	pieces := modelProto.GetPieces()
+	for _, p := range pieces {
+		v.Tokens = append(v.Tokens, p.GetPiece())
+		v.Scores = append(v.Scores, p.GetScore())
+		t := p.GetType()
+		switch t {
+		case sentencepiece.ModelProto_SentencePiece_UNKNOWN:
+		case sentencepiece.ModelProto_SentencePiece_CONTROL:
+		case sentencepiece.ModelProto_SentencePiece_UNUSED:
+		case sentencepiece.ModelProto_SentencePiece_BYTE:
+		default:
+			t = sentencepiece.ModelProto_SentencePiece_NORMAL
+		}
+		v.Types = append(v.Types, int32(t))
+	}
+
+	slog.Info(fmt.Sprintf("vocab size: %d", len(v.Tokens)))
+
+	// add any additional tokens
+	addIn, err := os.ReadFile(filepath.Join(dirpath, "added_tokens.json"))
+	if os.IsNotExist(err) {
+		return v, nil
+	} else if err != nil {
+		return nil, err
+	}
+
+	slog.Info("reading user defined tokens")
+
+	var extraTokenData map[string]int
+	if err := json.Unmarshal(addIn, &extraTokenData); err != nil {
+		return nil, err
+	}
+
+	type token struct {
+		key string
+		pos int
+	}
+
+	extraTokens := make([]token, 0)
+	for k, id := range extraTokenData {
+		extraTokens = append(extraTokens, token{k, id})
+	}
+
+	slices.SortFunc(extraTokens, func(a, b token) int {
+		return cmp.Compare(a.pos, b.pos)
+	})
+
+	numToks := len(v.Tokens)
+
+	for cnt, t := range extraTokens {
+		// the token id should match the specific index for the total number of tokens
+		if t.pos != cnt+numToks {
+			return nil, fmt.Errorf("token ID '%d' for '%s' doesn't match total token size", t.pos, t.key)
+		}
+		v.Tokens = append(v.Tokens, t.key)
+		v.Scores = append(v.Scores, -1000.0)
+		v.Types = append(v.Types, int32(llm.GGUFTokenUserDefined))
+	}
+	slog.Info(fmt.Sprintf("vocab size w/ extra tokens: %d", len(v.Tokens)))
+
+	if params.VocabSize > len(v.Tokens) {
+		missingTokens := params.VocabSize - len(v.Tokens)
+		slog.Warn(fmt.Sprintf("vocab is missing %d tokens", missingTokens))
+		for cnt := 0; cnt < missingTokens; cnt++ {
+			v.Tokens = append(v.Tokens, fmt.Sprintf("<dummy%05d>", cnt+1))
+			v.Scores = append(v.Scores, -1)
+			v.Types = append(v.Types, int32(llm.GGUFTokenUserDefined))
+		}
+	}
+
+	return v, nil
+}