Instruction file imported from Bryan-Roe-ai/Aria (.github/instructions/functions.instructions.md). Copyright stays with the author.

Azure Functions – function_app.py

• Endpoints: /api/chat, /api/chat-web, /api/tts, /api/quantum/*, /api/ai/status.
  • Verify runtime health at /api/ai/status (active provider, env vars, SQL pool, Cosmos status).
• Local dev:
  • Use the VS Code task: func: host start (depends on pip install (functions)).
  • Azurite storage emulator databases are present at repo root; Functions work offline.
• TTS configuration:
  • Azure Speech: set AZURE_SPEECH_KEY and AZURE_SPEECH_REGION (in env or local.settings.json).
  • Local fallback: set QAI_ENABLE_LOCAL_TTS=true; server will try Azure → pyttsx3 → gTTS.
• Chat provider detection (see shared/chat_providers.py): Azure OpenAI → OpenAI → LoRA → Local.
  • Azure requires: AZURE_OPENAI_API_KEY, AZURE_OPENAI_ENDPOINT, AZURE_OPENAI_DEPLOYMENT, AZURE_OPENAI_API_VERSION.
• Streaming responses (SSE): emit data: {json} lines and a final data: [DONE]; clients must parse SSE.
• Secrets: do not hardcode; prefer local.settings.json for dev or Azure App Settings in prod.
• Observability: Application Insights via shared/telemetry.py; failures are non-blocking.
• Optional Cosmos persistence: feature-flagged in shared/cosmos_client.py.
  • Recommended keys: QAI_ENABLE_COSMOS, COSMOS_ENDPOINT, COSMOS_KEY, COSMOS_DATABASE, COSMOS_CONTAINER.
  • Partition key: /session_id; enable TTL for ephemeral messages.
• Testing: prefer python .\\scripts\\test_runner.py --all; markers not slow and not azure for local.