New ai proxy

.github/actions/sentry_cli/action.yaml

@@ -0,0 +1,11 @@
+inputs:
+  version:
+    required: false
+    default: "2.39.1"
+runs:
+  using: "composite"
+  steps:
+    - run: curl -sL https://sentry.io/get-cli/ | SENTRY_CLI_VERSION="${{ inputs.version }}" sh
+      shell: bash
+    - run: sentry-cli --version
+      shell: bash

.github/workflows/ai_cd.yaml

@@ -0,0 +1,93 @@
+on:
+  workflow_dispatch:
+
+jobs:
+  compute-version:
+    runs-on: ubuntu-latest
+    outputs:
+      version: ${{ steps.version.outputs.version }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          fetch-tags: true
+      - run: git fetch --tags --force
+      - uses: ./.github/actions/doxxer_install
+      - id: version
+        run: |
+          VERSION=$(doxxer --config doxxer.ai.toml next patch)
+          echo "version=$VERSION" >> $GITHUB_OUTPUT
+          echo "Computed version: $VERSION"
+
+  build:
+    needs: compute-version
+    runs-on: depot-ubuntu-24.04-8
+    timeout-minutes: 60
+    steps:
+      - uses: actions/checkout@v4
+      - uses: ./.github/actions/rust_install
+        with:
+          platform: linux
+      - run: |
+          cargo build --release -p ai
+          objcopy --only-keep-debug target/release/ai target/release/ai.debug
+          objcopy --strip-debug --strip-unneeded target/release/ai
+          objcopy --add-gnu-debuglink=target/release/ai.debug target/release/ai
+        env:
+          CARGO_PROFILE_RELEASE_DEBUG: "true"
+      - uses: actions/upload-artifact@v4
+        with:
+          name: ai-binary
+          path: target/release/ai
+          retention-days: 1
+      - uses: actions/upload-artifact@v4
+        with:
+          name: ai-debug
+          path: target/release/ai.debug
+          retention-days: 1
+
+  sentry-upload:
+    needs: [compute-version, build]
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/download-artifact@v4
+        with:
+          name: ai-debug
+          path: ./debug
+      - uses: ./.github/actions/sentry_cli
+      - run: sentry-cli debug-files upload --include-sources ./debug
+        env:
+          SENTRY_AUTH_TOKEN: ${{ secrets.SENTRY_AUTH_TOKEN }}
+          SENTRY_ORG: ${{ secrets.SENTRY_ORG }}
+          SENTRY_PROJECT: ${{ secrets.SENTRY_PROJECT }}
+
+  deploy:
+    needs: [compute-version, build, sentry-upload]
+    runs-on: depot-ubuntu-24.04-8
+    timeout-minutes: 60
+    concurrency: ai-fly-deploy
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/download-artifact@v4
+        with:
+          name: ai-binary
+          path: ./apps/ai/bin
+      - run: chmod +x ./apps/ai/bin/ai
+      - uses: superfly/flyctl-actions/setup-flyctl@master
+      - run: flyctl deploy --config apps/ai/fly.toml --dockerfile apps/ai/Dockerfile --remote-only -e APP_VERSION=${{ needs.compute-version.outputs.version }}
+        env:
+          FLY_API_TOKEN: ${{ secrets.FLY_API_TOKEN }}
+
+  tag:
+    needs: [compute-version, deploy]
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/checkout@v4
+      - uses: mathieudutour/[email protected]
+        with:
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+          custom_tag: ai_v${{ needs.compute-version.outputs.version }}
+          tag_prefix: ""

.github/workflows/ai_ci.yaml

@@ -0,0 +1,23 @@
+on:
+  workflow_dispatch:
+  push:
+    branches:
+      - main
+    paths:
+      - apps/ai/**
+      - crates/llm-proxy/**
+      - crates/transcribe-proxy/**
+  pull_request:
+    branches:
+      - main
+    paths:
+      - apps/ai/**
+      - crates/llm-proxy/**
+      - crates/transcribe-proxy/**
+jobs:
+  ci:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: ./.github/actions/rust_install
+      - run: cargo check -p ai

Cargo.toml

@@ -10,9 +10,9 @@ debug = false
 [workspace]
 resolver = "2"
 members = [
+  "apps/ai",
   "apps/desktop/src-tauri",
   "apps/granola",
-  "apps/stt",
   "crates/*",
   "owhisper/*",
   "plugins/*",
@@ -48,6 +48,7 @@ hypr-language = { path = "crates/language", package = "language" }
 hypr-llama = { path = "crates/llama", package = "llama" }
 hypr-llm = { path = "crates/llm", package = "llm" }
 hypr-llm-interface = { path = "crates/llm-interface", package = "llm-interface" }
+hypr-llm-proxy = { path = "crates/llm-proxy", package = "llm-proxy" }
 hypr-loops = { path = "crates/loops", package = "loops" }
 hypr-mac = { path = "crates/mac", package = "mac" }
 hypr-moonshine = { path = "crates/moonshine", package = "moonshine" }

apps/stt/Cargo.toml → apps/ai/Cargo.toml

@@ -1,13 +1,10 @@
 [package]
-name = "stt-server"
+name = "ai"
 version = "0.1.0"
 edition = "2024"
 
-[[bin]]
-name = "stt"
-path = "src/main.rs"
-
 [dependencies]
+hypr-llm-proxy = { workspace = true }
 hypr-transcribe-proxy = { workspace = true }
 owhisper-providers = { workspace = true }
 
@@ -24,4 +21,5 @@ url = { workspace = true }
 
 dotenvy = { workspace = true }
 jsonwebtoken = { workspace = true }
-sentry = { workspace = true }
+sentry = { workspace = true, features = ["tower", "tower-axum-matched-path", "tracing"] }
+tower = { workspace = true }

apps/ai/Dockerfile

@@ -0,0 +1,9 @@
+FROM debian:bookworm-slim
+
+RUN apt-get update && apt-get install -y ca-certificates && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+COPY bin/ai /app/ai
+
+EXPOSE 3000
+CMD ["/app/ai"]

apps/ai/fly.toml

@@ -0,0 +1,42 @@
+app = 'hyprnote-ai'
+primary_region = 'sjc'
+kill_signal = 'SIGTERM'
+kill_timeout = 30
+swap_size_mb = 512
+
+[build]
+dockerfile = "Dockerfile"
+
+[deploy]
+strategy = "bluegreen"
+
+[env]
+PORT = "3000"
+SENTRY_ENVIRONMENT = "production"
+
+[http_service]
+processes = ['app']
+internal_port = 3000
+force_https = true
+auto_stop_machines = 'stop'
+auto_start_machines = true
+min_machines_running = 1
+
+[http_service.concurrency]
+type = "connections"
+hard_limit = 200
+soft_limit = 150
+
+[[http_service.checks]]
+grace_period = "20s"
+interval = "15s"
+method = "GET"
+path = "/llm/completions"
+protocol = "http"
+timeout = "4s"
+
+[[vm]]
+processes = ['app']
+memory = '1gb'
+cpu_kind = 'shared'
+cpus = 1