Skip to main content
Manifest is an open-source LLM router for agents and AI apps. Connect the providers you already use — subscriptions like ChatGPT or Claude, pay-per-token API keys, local models, and custom endpoints — and route every request through one OpenAI- and Anthropic-compatible URL. You choose which model handles each request. Add fallbacks so a rate limit or a retired model returns a working response instead of an error, cap spending per agent, and see the cost of every call.

​
Key features

All your providers

Connect API keys, subscriptions you already pay for, local models, or any custom endpoint.

LLM Gateway

Route each request to the model you pick, with fallbacks when one fails and limits that cap what it can spend.

Auto-fix

Repair a malformed request and resend it, instead of failing on something a retry can’t fix.

Observability

See what every agent spent, which models served it, and what failed.

​
Cloud or self-hosted?

Manifest is open source and runs two ways. Both speak the same API, so pointing a client at one or the other is a URL change.

Manifest Cloud

Runs on our servers. Sign up, connect a provider, and you’re routing. Recommended if you’re starting out.

Self-hosting

Runs on infrastructure you control: Docker on your own machine, a managed platform like Railway, your AWS or GCP account, or a server panel.
On the cloud there is nothing to install and nothing to configure on a server. You sign up, connect your providers, point your harness at the gateway URL, and everything else — routing, limits, Auto-fix, alerts — is a setting in the dashboard. The environment variables throughout these docs apply only to a Manifest instance you run yourself. Cloud is also subject to plan limits and can’t reach local models running on your machine. A self-hosted instance has neither restriction.