Genstack SDK Host is a FastAPI-based backend service that acts as a secure, multi-provider LLM (Large Language Model) gateway. It enables unified, authenticated, and auditable access to models from OpenAI, Google Gemini, and Anthropic, with robust tracking, encryption, and extensibility.

• Features
• Architecture
• API Reference
• Request & Response Schemas
• Authentication & Security
• Usage
• Development
• Testing
• Deployment
• Contributing
• License

• Unified LLM Gateway: Route requests to OpenAI, Google Gemini, and Anthropic models via a single API.
• API Key Authentication: Secure endpoints with per-user/project API keys.
• Encrypted Provider Keys: Provider credentials are encrypted using Fernet.
• Track & Audit: Per-model usage and error tracking, with periodic sync to Firestore.
• Extensible: Easily add new providers or models.
• Type-Safe & Modern: Built with FastAPI, async/await, and type hints.
• Test & CI Friendly: Designed for robust testing and CI/CD.

flowchart TD
    Client["Client (SDK/HTTP)"]
    API["FastAPI Server"]
    Router["Router (Provider Logic)"]
    Redis["Upstash Redis (Cache/Keys/Tracks)"]
    Providers["LLM Providers (OpenAI, Gemini, Anthropic)"]
    Firestore["Firestore (Persistent Usage Tracking)"]

    Client -->|HTTP POST /api/v1/sdk/generate| API
    API --> Router
    Router --> Redis
    Router -->|API Call| Providers
    Router --> Firestore

• Description: Generate a response from a specified LLM provider/model.
• Headers: x-api-key: <your-api-key>
• Body: See Request Schema
• Response: See Response Schema

• Description: Not allowed. Returns 405 with a message to use POST.

Example (app/schemas/request.json):

{
  "model": "gemini-2-5-pro-ggl",
  "config": {
    "streaming": false,
    "think": true,
    "tools": [""]
  },
  "messages": [
    {"role": "user", "content": "Hi! How are you doing?"}
  ]
}

Example (app/schemas/response.json):

{
  "data": "something"
}

Note: Actual response structure may vary by provider and model. See Formatter for details.

• API Key: All requests require a valid x-api-key header.
• Provider Keys: Stored encrypted using Fernet (FERNET_KEY in .env).
• Track Data: Usage and error counts are tracked per model and periodically flushed to Firestore.

• Python 3.10+
• Redis (Upstash or compatible)
• Google Firestore
• Provider API keys (OpenAI, Google Gemini, etc.)
• .env file with all required secrets (see below)

# Install dependencies
python -m venv env
env\Scripts\activate
pip install -r requirements.txt

# Set up .env (see below)
# Run the server
runserver.bat

curl -X POST http://localhost:8000/api/v1/sdk/generate \
  -H "x-api-key: <your-api-key>" \
  -H "Content-Type: application/json" \
  -d @app/schemas/request.json

Create a .env file in the project root with:

FERNET_KEY=...
UPSTASH_PROVIDERS_URL=...
UPSTASH_PROVIDERS_TOKEN=...
UPSTASH_KEYS_URL=...
UPSTASH_KEYS_TOKEN=...
UPSTASH_TRACKS_URL=...
UPSTASH_TRACKS_TOKEN=...
OPENAI_API_KEY=...
GEMINI_API_KEY=...
# Firebase credentials are loaded from app/firebase/firebase-adminsdk.json

• Run Locally: uvicorn app.main:app --reload
• Lint: ruff .
• Type Check: mypy .
• Tests: See tests/ and app/test/

• Example test scripts for OpenAI and Gemini are in app/test/.
• To run all tests: pytest

• Docker: (Dockerfile is present, but currently empty. Add your build steps as needed.)
• Production: Use a production ASGI server (e.g., uvicorn or gunicorn with uvicorn.workers.UvicornWorker).

Contributions are welcome! Please open issues or pull requests.

MIT (or specify your license here)