← Back to Discover

Bifrost

Open-source AI gateway that puts one OpenAI-style endpoint in front of many model providers, with fallback, budgets, virtual keys, MCP tool governance and built-in observability.

Model APIsMCP servers
Visit website ↗

What it does

Bifrost, from Maxim AI, is an LLM and MCP gateway that sits between your applications and model providers. You change the base URL of an existing SDK such as OpenAI, Anthropic, LiteLLM, Google GenAI or LangChain, and requests flow through Bifrost instead, so switching models does not require rewriting client code. The open-source edition includes a model catalog across providers (the site cites 8+ providers and 1,000+ models, plus custom deployed ones), automatic fallback between providers, spending limits, virtual keys with independent budgets and access control, an MCP gateway that centralizes tool connections and policy, and OpenTelemetry-based observability with a built-in dashboard. A separate Enterprise edition is offered for additional features and support, with a 14-day free trial. Bifrost can be deployed inside your own VPC, and the site says no data passes through Maxim's servers in that setup. You can start locally with a single npx command.

Capabilities

  • One OpenAI-style endpoint for many providers and models
  • Drop-in base URL change for OpenAI, Anthropic, LiteLLM, Google GenAI and LangChain SDKs
  • Automatic fallback between providers
  • Budgets and virtual keys for teams and use cases
  • Built-in MCP gateway for tool connections and policy
  • OpenTelemetry observability with a built-in dashboard

When to use it

  • Routing an app across several model providers without changing client code
  • Setting spending limits and separate access keys for teams or projects
  • Managing MCP tool access for agents in one central place
  • Moving from LiteLLM to a self-hosted gateway

Alternatives to explore

  • LiteLLM