What it does
Manifest is an open-source gateway that sits between your agents or apps and the model providers they call. The site organises it into three steps: route all requests through one endpoint to your models and providers, fix problems by falling back when a provider is down and repairing broken requests, and observe every request, token and dollar spent. It works with OpenAI-compatible clients, your own keys, subscription logins, custom providers and local models such as Ollama, LM Studio and llama.cpp, so you decide what goes to the cloud and what stays on your machine. Autofix is the headline feature. The site describes an agent that repairs a failing request before the error comes back to your application, available through the gateway or on its own. Other features include model fallbacks, full request and response body logs, visual setup of model parameters per route, and cost breakdowns by agent, key and provider. Manifest is offered as a cloud version for easy onboarding or a self-hosted Docker version. The site also points to a catalog of common LLM API errors, a community catalog of model parameters, and a list of LLM APIs with free text tiers. The team offers custom development around the gateway.
Capabilities
- Single endpoint for many providers, keys, subscriptions and local models
- Autofix that repairs broken requests before your agent sees the error
- Model fallbacks when a provider goes down
- Full request and response body logs
- Cost visualisation by agent, key and provider
- Self-host with Docker or use the cloud version
When to use it
- Your agents hit provider outages or avoidable request errors and you want automatic recovery
- You want one place to watch token usage and spend across providers
- You want to combine your own API keys, subscriptions and local models
- You prefer an open-source gateway you can run on your own infrastructure
