Lattice
One OpenAI-compatible API fronting five providers. The gateway holds provider credentials, decides which one handles a request, retries a fallback when one fails, and records what happened — without any of that bookkeeping blocking the response a client sees.
Application A └ Application B ├──▸ Lattice ▸── OpenAI / Anthropic / Google / Ollama / vLLM Application C ┘
Why Lattice
04 CapabilitiesOne API
A single /v1/chat/completions endpoint fronts five providers. Swap or add a provider without touching client code.
Failover
Five consecutive failures opens a per-provider circuit; requests route around it until a 30-second cooldown elapses.
Policy
Model allowlists, monthly token quotas, per-minute rate limits, and routing strategy are enforced on every request.
Observability
Every request is logged with provider, latency, attempts, and estimated cost — queryable through the dashboard.
Documentation
8 SectionsGetting Started
Install, seed keys, first request.
Architecture
Request lifecycle, routing, security.
Providers
Cloud and self-hosted backends.
API Reference
Auth, completions, errors, limits.
Management API
Applications, keys, usage.
Dashboard
Setup, auth, feature tour.
Development
Standards, testing, packages.
Deployment
Configuration and production guide.