Flit Documentation
Welcome to the comprehensive documentation hub for Flit, the enterprise-grade AI gateway built on LiteLLM. Explore guides, architectural specifications, API references, and deployment best practices.
Core Documentation Sections
Spin up Flit locally with Docker Compose in under five minutes, initialize administrator credentials, and make your first routed completion request.
Understand the separation between the high-throughput inference Data Plane and the authoritative Control Plane, zero-IO memory routing, and durable state.
Configure aquila auto-classification, plus aquila-fast, aquila-smart, and aquila-power tiers. Set up Regex patterns, LLM classifier prompts, and zero-IO hybrid fallback logic.
OpenAI-compatible specifications for /v1/chat/completions, /v1/responses, /v1/models, native MCP Streamable HTTP (/mcp), and A2A agent endpoints.
Protect corporate boundaries with Microsoft Presidio PII anonymization/blocking, Microsoft Entra ID (OIDC) SSO, and encrypted session cookies.
Partition developers into teams, issue scoped virtual keys with rate limits and spend caps, and govern access across models, MCP tools, and A2A agents.
System Requirements
To run Flit in your environment, ensure you meet the following baseline requirements:
- Docker & Docker Compose: Compose v2 or higher.
- Operating System: Linux (Ubuntu, Debian, RHEL) or macOS (Apple Silicon / Intel).
- Database: PostgreSQL 14+ (provided out-of-the-box in local Compose).
- Cache & State: Redis 7+ (provided out-of-the-box in local Compose).
- Client Runtime: Any HTTP client, Python 3.9+, Node.js 18+, Go, or curl.