Bitrouter is an open-source, context-aware model router built with Rust, designed for agentic workflows. It functions as an MCP server, acting as an LLM router and local proxy. Bitrouter connects to various LLM providers and OpenAI-compatible endpoints, serving as both an MCP and ACP gateway. Developers and teams building AI coding agents can use Bitrouter to manage their LLM interactions. It helps refine routing decisions based on live production traffic, supporting adaptive policy routing. This tool aims to enhance the efficiency and accuracy of LLM calls in production environments.

What does Bitrouter connect to?

Bitrouter acts as a context-aware LLM router and local proxy, connecting to many Large Language Model providers. These include OpenAI, Anthropic, Google, Groq, DeepSeek, Mistral, Moonshot, MiniMax, and Nvidia. It also supports any OpenAI-compatible endpoint. Functioning as an MCP server, Bitrouter serves as both a Model Context Protocol gateway and an Agent Client Protocol gateway, enabling communication across diverse AI services.

Core features and tools

Bitrouter exposes several callable tools. As an MCP client, it offers complete, list_models, and status. Built-in server tools include web_fetch and web_search. The /bitrouter Agent Skill assists AI coding agents with installation, configuration, migration, and troubleshooting. Its command-line interface provides operational commands like "serve", "start", "stop", "status", "restart", "reload", "wallet", "key", and "policy".

Who is it for?

Bitrouter targets developers and teams building AI coding agents and agentic workflows. Its purpose is to optimize the efficiency and accuracy of LLM calls in production. It helps manage provider selection and routing decisions for various AI tasks. This tool is valuable for those seeking to refine how their agents interact with LLMs. Skip this if you are not developing AI coding agents or managing agentic workflows that involve multiple LLM providers.

How to get started

The server authenticates using an API key passed as an environment variable or through Bitrouter Cloud sign-in. For Cloud authentication, access is granted through a browser OAuth device flow or an API key for non-interactive login.

Use cases

• Routing each call, tool, and agent to the most efficient path that still achieves the goal, dynamically refining this routing during the workflow.
• Continuously improving agent workflows by allowing users to send "bitrouter/auto" instead of a specific model name, enabling a policy to adapt routing decisions based on live production traffic.
• Rerouting requests mid-run in case of rate limits to prevent re-paying for completed work, and optimizing provider selection based on efficiency, latency, or throughput for each call.

Bitrouter provides a flexible routing and proxy solution for managing diverse LLM interactions within agentic workflows. It supports multiple providers and offers tools for operational control and agent assistance. The platform aims to enhance the performance of AI coding agents through adaptive policies and efficient request handling.