Microservice for AI workflow blocks. Provides OpenRouter and OpenAI with rate limiting, retries, JSON schema validation. Backend calls it for AI Transform nodes; API keys never touch the frontend.

llm-service/
├── src/
│   ├── config/        # Load and validate config
│   ├── providers/     # OpenAI, OpenRouter
│   ├── services/      # Chat service
│   ├── types/         # Contracts, error codes
│   ├── utils/         # HMAC, rate limiter, concurrency
│   ├── server.ts      # Fastify server
│   └── index.ts       # Entry point
├── config/models.json
└── package.json

Prerequisites: Node.js 20+, OPENROUTER_API_KEY or OPENAI_API_KEY

npm install
cp .env.example .env
# Edit .env: HMAC_SECRET (must match backend), API keys (see Environment variables below)

npm run build && npm run start

Dev with hot reload: npm run dev — runs on port 3002

Health (no auth): GET /health, GET /ready

Authenticated (HMAC: x-timestamp, x-signature): GET /v1/models, POST /v1/chat

POST /v1/chat body: provider (openai | openrouter), model, messages, temperature, maxOutputTokens, responseSchema, requestId, userId.

Backend must set LLM_SERVICE_BASE_URL and matching HMAC_SECRET.

INVALID_REQUEST, MODEL_NOT_FOUND, LLM_MODEL_NOT_CONFIGURED, PROVIDER_NOT_CONFIGURED, RATE_LIMIT_EXCEEDED, CONCURRENCY_LIMIT_EXCEEDED, PROVIDER_ERROR, PROVIDER_TIMEOUT, JSON_VALIDATION_FAILED, INTERNAL_ERROR

MIT License