Introduction
Lattice is centralized AI infrastructure: an internal gateway that sits between company applications and the model providers (OpenAI, Anthropic, Google, Ollama, vLLM) they'd otherwise integrate with one by one. Instead of every application shipping its own provider SDKs, API keys, retry logic, and cost tracking, applications talk to Lattice once, and Lattice talks to the providers.
The problem it solves
Without a gateway, every application that wants to call an LLM ends up re-solving the same problems: where do provider credentials live, what happens when a provider has an outage, how do you know which model an "smart" alias should resolve to this month, and who is spending how many tokens. Lattice centralizes all of that in one place that's operated once and consumed many times.
Two applications, one repository
Lattice is a two-app monorepo:
apps/gateway— a Nitro + h3 server. This is the thing applications and providers actually talk to: the OpenAI-compatible/v1API, the/management/v1admin API, authentication, quota/rate-limit enforcement, routing, and request logging.apps/dashboard— a Nuxt 4 + Nuxt UI admin surface, built entirely on top of the gateway's own management API. It has no privileged access the management API doesn't already grant; it just makes that API usable by a human.
Who this documentation is for
Next steps
Continue to Installation to set up the repository, or jump straight to Quick Start if you already have Bun and Postgres ready.