Manifest

Manifest is an open-source model router for AI agents and apps. It provides a single OpenAI-compatible endpoint that connects to multiple LLM providers, including API keys, subscriptions, local models, and custom providers. Each query is automatically routed to the best model based on complexity, specificity, and custom HTTP headers. Key features include intelligent routing based on request complexity and model capability, cost tracking across all providers with notifications and spending limits, automatic fallback to alternative models when a query fails, and support for mixing API keys, subscriptions, and local models through one unified interface. Manifest ships as a Docker image for easy self-hosting with a single command. With 7,200+ GitHub stars and an MIT license, Manifest is ideal for AI developers and teams looking to reduce provider lock-in, optimize LLM costs, and add resilience to their AI agent pipelines. The web UI provides full visibility into routing decisions, usage patterns, and spending across all connected providers.