What Is Manifest (manifest.build)?

Manifest is an open-source LLM gateway that routes your AI agents and agentic workflows to many model providers through a single endpoint. It fits teams that want one place to connect subscription plans, pay-per-token APIs, local models, and custom providers, while adding reliability features like automatic fallbacks and request repair. It is aimed at AI builders, teams and startups, and platforms or providers who need to manage inference consumption and keep agents running when individual providers fail.

What the gateway actually does

The core mechanism is routing. Instead of wiring each agent directly to each provider, you point your agent at Manifest's single endpoint and let it decide which model or provider handles a request. The site describes three primary functions:

• Route — one endpoint to access all your models and providers.
• Fix — fallback when providers are down, and repair broken requests on the fly.
• Observe — track every request, token, and dollar spent.

Manifest is OpenAI-compatible, so existing tooling built around the OpenAI API format can generally point at it. It supports bringing your own key (BYOK), OAuth tokens for subscription plans, custom providers, and local models.

Autofix and fallbacks

Manifest's headline reliability feature is Autofix. The site cites Datadog's State of AI Engineering for the claim that 5% of LLM API calls fail in production and that most of those errors are avoidable. Autofix is described as an agent that repairs a bad request before the error reaches your application, available through the LLM Gateway or standalone. Separately, model fallbacks automatically retry on another provider when one goes down.

A concrete example: if an agent sends a malformed request or a provider returns an error, Manifest attempts to repair or reroute it so your workflow sees a successful response instead of a failure. That matters most for production agents where a single failed call can stall a multi-step task.

Observability and cost control

Manifest logs the full request and response body for every call, and visualizes cost broken down by agent, key, and provider. Model and parameters can be set per route visually, without changing code. These features target the operational side of running agents: knowing what was sent, what came back, and what it cost.

Deployment and model options

The self-hosted router is presented as free and open source. Pricing for the cloud option is not specified in the available material, so check the pricing page before assuming cost or access terms.

Who it is for

• AI builders — connect subscription plans, BYOK providers, and local models to your harness quickly, and customize routing to maximize existing plans.
• Teams and startups — manage inference consumption and settings, and track requests and tokens through an interface.
• Platforms and providers — reduce avoidable client errors that hurt user experience and revenue, and help users connect smoothly to models.

How to decide if it fits

Manifest is a reasonable fit if you are running agents or workflows across more than one provider, want a single OpenAI-compatible endpoint, and care about surviving provider outages and malformed requests without building that logic yourself. Self-hosting and local-model routing also suit teams with data-residency or cost constraints.

It is less clearly necessary if you use a single provider, have no production reliability concerns, or already have a mature internal gateway. In those cases the added routing layer may not earn its place. Before committing, confirm two things on the site: the current cloud pricing and whether the specific providers and subscription plans you rely on are supported.