<- all tokdocs

ClawRouter: A Smart LLM Router That Cuts Your Inference Costs

Watch on TikTok

View on TikTok ->

ClawRouter is an open-source LLM router that automatically sends each request to the cheapest model capable of handling it. Simple questions go to cheap models. Complex code goes to expensive ones. It runs locally, supports micropayments so you do not need to manage API keys for every provider, and claims to cut inference bills by around 80%.

How the Routing Works

ClawRouter uses a 14-dimension weighted scoring system that runs entirely on your machine in under 1 millisecond. No external API calls are made for the routing decision itself.

The router classifies requests into four tiers, each mapped to a different model:

Tier Primary Model Cost/M Savings vs Opus
SIMPLE deepseek-chat $0.27 99.6%
MEDIUM gpt-4o-mini $0.60 99.2%
COMPLEX claude-sonnet-4 $15.00 80%
REASONING o3 $10.00 87%

GitHub README showing ClawRouter's tier-to-model mapping, cost savings tables, and supported models across OpenAI, Anthropic, Google, and other providers

Based on traffic distribution (45% simple, 35% medium, 15% complex, 5% reasoning), the blended average cost comes to $3.17/M tokens, compared to $75/M for Claude Opus. That is a 96% savings on a typical workload.

Setup

Installation takes three steps:

  1. Install the plugin: openclaw plugin install @blockrun/clawrouter
  2. Fund your wallet with USDC on Base (the address prints on install). $5 is enough for thousands of requests.
  3. Enable smart routing: openclaw config set model blockrun/auto

GitHub README showing the ClawRouter overview with the tagline "Save 78% on LLM costs. Automatically" and the quick start installation steps

Every request now routes to the cheapest capable model. The system supports 30+ models across OpenAI, Anthropic, Google, DeepSeek, xAI, and Moonshot through a single wallet.

Key Features

  • 100% local routing. The scoring runs on your machine. No data leaves for routing decisions.
  • Zero external calls for routing. No API calls to decide which model handles the request.
  • x402 micropayments. Pay per request with USDC on Base. No need to manage API keys for each provider.
  • Open source. MIT licensed with fully inspectable routing logic.

Detailed view of the tier-to-model mapping and cost savings tables, showing the pricing for each supported model across providers

Key Takeaways

  • ClawRouter routes each LLM request to the cheapest model that can handle it, using local scoring with no external API calls
  • The four-tier system (simple, medium, complex, reasoning) maps to different models at different price points
  • Setup is three commands and a $5 USDC deposit, then every request auto-routes through a single wallet

Resources

Published May 25, 2026. Writeup generated from a favorited TikTok.