BlockRunAI/ClawRouter

What it solves

ClawRouter addresses the difficulty autonomous AI agents have in accessing paid LLM APIs, as agents cannot sign up for accounts or provide credit cards. It provides a way for agents to operate independently by allowing them to pay for LLM usage via wallet signatures and USDC micropayments, while also offering a traditional credit card payment path for human users.

How it works

ClawRouter acts as a local proxy that runs on the user's machine. It uses a 15-dimension scoring system to analyze requests locally in under 1ms and route them to the cheapest capable model among 76 available models from providers like OpenAI, Anthropic, and Google.

Authentication is handled either through a credit-card-funded API key or a non-custodial wallet signature using the x402 protocol for USDC micropayments on Solana or Base. The routing logic is powered by @blockrun/router-core, which applies hard filters (such as tool use, vision, and context window) to find the best model for the task.

Who it’s for

• Autonomous AI agents that need a way to pay for their own API costs independently.
• Developers using OpenAI-compatible clients (like Cursor, continue.dev, or VS Code) who want to reduce LLM costs through smart routing.
• Users who want access to a wide variety of models with a single endpoint and a free tier of 7 open-weight models.

Highlights

• Agent-native billing: Supports USDC micropayments via x402, removing the need for accounts or credit cards for agents.
• Smart local routing: Reduces costs by up to 84% by routing requests to the cheapest capable model locally with minimal latency.
• Broad model support: Access to 76 chat models across 9 providers behind a single endpoint.
• Open source: MIT licensed and runs locally as a proxy.
• Flexible payment: Supports credit cards, USDC, or a free tier for specific models.