Manifest LLM Gateway

What it does

Manifest LLM Gateway sits between an AI client and model providers: point the client at the gateway and send requests through it. The gateway authenticates the request, applies configured limits, selects a model using a default or header-matched tier, and forwards the request with provider credentials. Clients can use the `auto` model value for configured routing or specify a model ID directly. Responses are streamed back to the client.

Builders and teams use the gateway to manage model access and inference consumption for agents and AI applications. It supports fallback models when provider requests fail, and its Autofix feature may repair and resend a failing request before fallback. Configuration and usage management are available through a dashboard; the project also documents self-hosting with Docker. The gateway is OpenAI-compatible and can connect to provider API keys, subscription-based providers, custom providers, and local models.

Use requires configuring a gateway account or self-hosted instance, connecting a provider, and directing a compatible client to the gateway with a Manifest key. Provider access and any provider-side usage charges remain separate considerations. Fallback requires multiple available routes to provide meaningful provider redundancy; with one provider it can only switch between that provider’s models.

What it can do

• Route requests to models using default or header-matched custom tiers
• Retry failed requests across configured fallback models
• Repair and resend some failing requests with Autofix
• Apply per-harness request limits
• Log and inspect request and response bodies
• Track inference usage and costs by agent, key, and provider

Works with

• OpenAI-compatible clients
• Anthropic Messages API
• Provider API keys
• Subscription-based providers
• Custom providers
• Local model servers including Ollama, LM Studio, and llama.cpp

Before you choose

• Provider credentials and provider-side inference charges are separate; the site describes bringing your own key and paying usage directly.
• Fallback needs more than one available provider for provider-level redundancy; with one provider it can only switch among that provider’s models.
• Self-hosted users must configure and operate their own deployment; vendor documentation does not specify supported operating systems.

Listed under

• Coding
• MCP and agent tools
• API-proxy
• Developers
• LLM-gateway
• LLM-routing
• Model-fallback
• Open-source

Self-hosting is free and open source. A cloud Free plan is documented, but its allowances and paid cloud rates require confirmation. Bring model-provider access and budget for its usage and operating infrastructure.

We couldn’t find published prices for Manifest LLM Gateway. See their pricing page ↗