# ClawRouter: An LLM Router Built for Autonomous Agents

> ClawRouter is an MIT-licensed TypeScript LLM router that routes each request to the cheapest capable model in under 1ms using 15-dimension local scoring, and uniquely supports agent-native payments via USDC micropayments over the x402 protocol so that autonomous agents can pay for inference without a human account.

**BlockRunAI/ClawRouter** — The agent-native LLM router for autonomous agents. Every frontier model behind one wallet, <1ms local routing, USDC payments on Base & Solana via x402.

- Repository: https://github.com/BlockRunAI/ClawRouter
- Website: https://www.npmjs.com/package/@blockrun/clawrouter
- Stars: 6,614 · Forks: 653
- Language: TypeScript
- License: MIT
- Published: 2026-09-09 · Updated: 2026-09-09 · Language: en
- Canonical page: https://hysenlabs.com/projects/blockrunai-clawrouter

## Why Autonomous Agents Need a Different Kind of LLM Router

Most LLM routing products assume a human developer is managing the account. They require signing up for a dashboard, entering a credit card, obtaining an API key, and configuring the key in a secrets manager. An autonomous agent cannot do any of those things.

ClawRouter is built around the observation that agents can sign cryptographic transactions. The x402 protocol allows an agent to pay for an inference request by signing a USDC micropayment on Solana or Base before sending the request. No account creation, no credit card, no API key. The wallet is generated locally when the CLI starts for the first time. The README states: "Agents can't sign up for accounts. Agents can't enter credit cards. Agents can only sign transactions."

For human developers who prefer credit cards, the router also takes an API key generated at user.blockrun.ai. The same router binary, the same 79 models, and the same local routing logic apply either way. The payment mechanism is a configuration detail rather than a separate product.

## How the 15-Dimension Local Routing Works

ClawRouter analyzes each incoming request across 15 dimensions before routing it to a model. The routing decision happens entirely locally in under 1ms, without sending the request to an external service to decide where it should go. The open-source router-core repository contains the scoring and constraint-first ranking algorithm.

The 15 dimensions are not fully enumerated in the README, but the described behavior shows the router considers: whether the request requires a specific context length, whether it is a reasoning task or a coding task, whether the model needs to support specific output formats, and the cost per token of each available model. The router finds the cheapest model that satisfies all the constraints for that request.

The README states that on a published workload mix, this routing approach costs 84% less than pinning Claude Opus 5 for all requests. The comparison is explicitly described as computed from that published workload mix rather than an estimate.

The catalog covers 79 chat models from OpenAI, Anthropic, Google, xAI, DeepSeek, and more, plus 12 image models, 8 video models, 5 voice models, and additional specialized endpoints for web search, financial data, and on-chain queries.

## Getting Started with ClawRouter

ClawRouter is published as the npm package @blockrun/clawrouter. The package.json shows the binary is registered as `clawrouter`, pointing to `dist/cli.js`. The current version is 0.12.279.

For a credit-card-funded workflow, sign up at user.blockrun.ai, top up your balance, mint an API key, and run:

```bash
clawrouter login brk_live_…
```

For the agent-native USDC workflow, a wallet is generated locally on first run with no signup required. No command is needed beyond starting the CLI.

The free tier includes 6 models with no signup and no wallet. The README describes the free models as including a 1M-context reasoner and two sub-second coders.

The package exports a router entry point at `@blockrun/clawrouter/router` for embedding ClawRouter's routing logic directly into a TypeScript or JavaScript application, in addition to the CLI interface. The package includes skills/ and docs/ directories with additional configuration documentation.

## Payment Paths: USDC, Credit Card, and Free Tier

ClawRouter provides three distinct ways to pay for inference.

The free tier gives access to 6 open-weight models with no account, no wallet, and no API key. These models include a 1M-context reasoner and two sub-second coders. The README notes that image generation turns route to paid models even on the free tier.

The USDC path uses the x402 protocol for per-request micropayments. A local wallet is generated on first run. The wallet address on Solana or Base holds USDC, and each inference request deducts the cost at the moment it is fulfilled. This is the path designed for autonomous agents: the agent holds USDC, signs each payment, and operates without any human-managed account.

The credit card path works through a BlockRun account. You fund the account at user.blockrun.ai, generate an API key with the `brk_live_` prefix, and the router sends requests to api.blockrun.ai using that key, billing your account balance. This path requires no crypto knowledge and works for teams that prefer traditional payment flows.

For organizations with compliance or regulatory constraints around cryptocurrency, the credit card path is a workable alternative. The routing logic and model catalog are identical across both paths.

## What ClawRouter Covers Beyond Chat Models

The README describes ClawRouter as a gateway to multiple resource types, not just language models. Beyond 79 chat models, the BlockRun catalog includes image generation, video generation, speech synthesis (5 voices), image editing, and outbound phone calls that return as transcripts.

Data sources are also available: web, news, and neural search; prediction market data; live crypto and equity quotes; on-chain SQL over more than 100 million labeled wallets; DEX routing; and RPC access across 40 chains.

For developers building autonomous agents that need access to real-world data as part of their inference pipeline, the availability of financial data and on-chain queries through the same router reduces the number of separate API integrations required. Each of these resources is billed per request using the same USDC or credit-card mechanism as LLM calls.

ClawRouter also provides an OpenClaw plugin interface (openclaw.plugin.json at the repository root) that allows the router to be used as an extension in compatible AI coding assistant environments.

## Limitations and Considerations

ClawRouter routes to the cheapest capable model, which means you are not always in direct control of which specific model handles a request. If a workflow requires consistent output format or behavior across requests, routing variability is a risk. The README describes a constraint-first ranking approach, but the specific 15 dimensions are not fully documented, which makes it harder to reason about why a particular model was chosen for a given request.

The x402 USDC payment path requires the agent to hold cryptocurrency and pay gas fees or transaction costs. For teams in jurisdictions where cryptocurrency payments raise compliance concerns, or for developers unfamiliar with managing wallets, this path adds friction.

The project is at version 0.12.x, which suggests it is not yet at a stable 1.0 API. The CHANGELOG.md shows frequent releases. Teams building production systems that depend on ClawRouter should pin a specific version and monitor breaking changes.

The free tier is limited to 6 models. Production use for anything beyond simple testing will incur per-request costs.

## ClawRouter versus OpenRouter

OpenRouter is the closest alternative: it also provides access to many LLM providers behind a single endpoint and a single billing account. The differences are meaningful.

OpenRouter requires an account and API key even for its free tier. It routes requests server-side. It does not support x402 USDC agent payments. Routing decisions in OpenRouter are configured by the caller specifying preferences rather than computed locally by a client-side algorithm.

ClawRouter's local routing runs the scoring algorithm on the requester's machine before the request leaves. This means routing latency is sub-1ms rather than adding a network round-trip. For autonomous agents, the x402 path is the distinguishing feature: no human account is needed at all.

For teams that need a simple multi-provider API key aggregator and prefer not to manage a local routing process, OpenRouter may be simpler. For teams building agents that need autonomous payment capability or for teams specifically optimizing cost by routing across providers, ClawRouter addresses those requirements directly.

## Conclusion

ClawRouter is the right fit for developers building autonomous AI agents that need to call LLMs without human-managed API keys, or for teams reducing inference costs by routing requests across multiple providers instead of pinning one model. The 84% cost reduction figure comes from a published workload mix comparison against pinning Claude Opus 5. The x402 USDC payment path is the meaningful differentiator from other LLM routers: it lets agents operate financially without a human managing an account. It is not the right choice for teams that want simple, direct access to one specific model without routing logic, or for teams that cannot use USDC or prefer not to manage a crypto wallet. Verify that your required models appear in the catalog at the @blockrun/clawrouter npm package page before committing to it.

## FAQ

### How to use claw router?

Install the package with npm install @blockrun/clawrouter. For credit-card billing, sign up at user.blockrun.ai, top up, mint an API key, and run clawrouter login brk_live_... to authenticate. For the USDC wallet path, a wallet is generated locally on first run with no signup required. Six models are available for free with no setup at all.

### What is claw router?

ClawRouter is an open-source LLM router that sends each request to the cheapest capable model using 15-dimension local scoring in under 1ms. It is the only LLM router that supports autonomous agent payments via USDC micropayments over the x402 protocol on Solana and Base, so agents can pay for inference without a human-managed account.

### Is claw router free?

Six models are permanently free with no signup, no API key, and no wallet required. Paid usage is billed per request either through a USDC wallet (x402 protocol) or through a credit-card-funded API key obtained at user.blockrun.ai.

### Claw router vs OpenRouter: what is the difference?

ClawRouter runs its routing algorithm locally in under 1ms before a request leaves the machine, whereas OpenRouter routes server-side. ClawRouter supports autonomous agent payments via USDC micropayments over x402 with no account needed. OpenRouter requires an account and API key even for its free tier and does not support x402 payments.

### What are the alternatives to ClawRouter?

OpenRouter is the closest alternative, providing multi-provider LLM access through a single endpoint. Unlike ClawRouter, it requires an account and routes server-side. LiteLLM is another option that provides a unified API over multiple providers with local routing logic, though it does not include the x402 USDC agent payment path.

## Sources

- [BlockRunAI/ClawRouter on GitHub](https://github.com/BlockRunAI/ClawRouter)
- [License: MIT](https://github.com/BlockRunAI/ClawRouter/blob/main/LICENSE)
- [Project website](https://www.npmjs.com/package/@blockrun/clawrouter)
- [README](https://github.com/BlockRunAI/ClawRouter/blob/main/README.md)
- [Releases](https://github.com/BlockRunAI/ClawRouter/releases)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/blockrunai-clawrouter