The LLM router built for autonomous agents

Agents can't sign up for accounts. Agents can't enter credit cards.
Agents can only sign transactions.

ClawRouter is the only LLM router that lets agents operate independently.

8 models free, no crypto required. No signup. No API key. No credit card.

         

ClawRouter is an open-source smart LLM router that reduces AI API costs by up to 92%. It analyzes each request across 15 dimensions and routes to the cheapest capable model in under 1ms, entirely locally. ClawRouter is the only LLM router built for autonomous AI agents — it uses wallet signatures for authentication (no API keys) and USDC micropayments via the x402 protocol (no credit cards). 55+ models from OpenAI, Anthropic, Google, xAI, DeepSeek, and more. MIT licensed.

Why ClawRouter exists

Every other LLM router was built for human developers — create an account, get an API key, pick a model from a dashboard, pay with a credit card.

Agents can't do any of that.

ClawRouter is built for the agent-first world:

• Starts at $0 — 8 NVIDIA models are free forever (incl. 675B Mistral Large 3, Qwen3.5 122B + a vision-capable Nemotron Omni)
• No accounts — a wallet is generated locally, no signup
• No API keys — your wallet signature IS authentication
• No model selection — 15-dimension scoring picks the right model automatically
• No credit cards — agents pay per-request with USDC via x402
• No trust required — runs locally, <1ms routing, zero external dependencies

This is the stack that lets agents operate autonomously: x402 + USDC + local routing.

How it compares

✓ Open source · ✓ Smart routing · ✓ Runs locally · ✓ Crypto native · ✓ Agent ready

We're the only one that checks all five boxes.

Quick Start

No wallet? 8 models work free out of the box. Install, run, and pin nvidia/gpt-oss-120b (or any of the 8) — no crypto, no signup, no balance required. Add USDC later when you want paid models.

Option A — OpenClaw Agent

OpenClaw is an AI coding agent. If you're using it, ClawRouter installs as a plugin. Two paths:

A1. Recommended — one-shot install script:

curl -fsSL https://blockrun.ai/ClawRouter-update | bash
openclaw gateway restart

This handles everything: registration, models config, auth profile, wallet setup. Smart routing (blockrun/auto) is now your default model.

A2. If you prefer pure npm:

npm install -g @blockrun/clawrouter
clawrouter setup            # finishes OpenClaw integration — REQUIRED
openclaw gateway restart

⚠️ Skipping clawrouter setup will leave you broken. Bare npm install -g only puts the package on disk; it does NOT register the plugin with OpenClaw, sync the models allowlist, or write the auth profile. Symptom: /models in your bot shows only ~7 entries (OpenClaw's hardcoded defaults) instead of the full ~38 BlockRun models. Run clawrouter setup to repair, or use path A1 to begin with.

Option B — Standalone (continue.dev, Cursor, VS Code, any OpenAI-compatible client)

Using Claude Code? Check out BRCC — it's purpose-built for Claude Code with the same smart routing and x402 payments.

Using NousResearch Hermes? See ClawRouter-Hermes — a Python plugin that wires Hermes into the ClawRouter proxy. Same wallet, same 55+ models, same x402 USDC settlement on Base & Solana.

No OpenClaw required. ClawRouter runs as a local proxy on port 8402.

1. Start the proxy

npx @blockrun/clawrouter

2. Fund your wallet — optional, skip for free tier Your wallet address is printed on first run. For paid models, send a few USDC on Base or Solana — $5 covers thousands of requests. To stay at $0, pin any of the 8 free models (e.g. nvidia/gpt-oss-120b) or use /model free inside OpenClaw.

3. Point your client at http://localhost:8402

models:
  - name: ClawRouter Auto
    provider: openai
    model: blockrun/auto
    apiBase: http://localhost:8402/v1/
    apiKey: x402
    roles:
      - chat
      - edit
      - apply

{
  "models": [
    {
      "title": "ClawRouter Auto",
      "provider": "openai",
      "model": "blockrun/auto",
      "apiBase": "http://localhost:8402/v1/",
      "apiKey": "x402"
    }
  ]
}

from openai import OpenAI
client = OpenAI(base_url="http://localhost:8402", api_key="x402")
response = client.chat.completions.create(model="blockrun/auto", messages=[...])

Routing Profiles

Choose your routing strategy with /model <profile>:

Shortcuts: /model grok, /model br-sonnet, /model gpt5, /model o3

How It Works

100% local routing. <1ms latency. Zero external API calls.

Request → Weighted Scorer (15 dimensions) → Tier → Best Model → Response

Blended average: $2.05/M vs $25/M for Claude Opus = 92% savings

Image Generation

Generate images directly from chat with /cr-imagegen:

/cr-imagegen a dog dancing on the beach
/cr-imagegen --model dall-e-3 a futuristic city at sunset
/cr-imagegen --model banana-pro --size 2048x2048 mountain landscape

The slash command is /cr-imagegen to avoid colliding with Telegram channel commands. Typing /imagegen in chat still works for backward compatibility.

Video Generation

Generate short AI videos directly from chat with /videogen:

/videogen a red apple slowly spinning
/videogen --model seedance-2-fast --duration=5 a cat waving
/videogen --model grok-video a neon city at night

Or drive it over HTTP — ClawRouter proxies the BlockRun gateway, handles x402 payment, and downloads the returned MP4 to local disk, rewriting url to http://localhost:8402/videos/<file>.mp4 so the asset survives past the upstream's temporary bucket.

curl -X POST http://localhost:8402/v1/videos/generations \
  -H "Content-Type: application/json" \
  -d '{"model":"bytedance/seedance-2.0-fast","prompt":"a red apple slowly spinning","duration_seconds":5}'

Seedance is token-priced upstream at ~20,256 tokens/sec — the blockrun videos route now defaults Seedance to resolution=720p with generate_audio=true for text-to-video (2× the per-second token count of the older 480p baseline; audio is included in that rate). The quote is duration × tokens/sec × $/1M tokens × 1.05 margin. Image-to-video is priced the same as text-to-video (the earlier ~40% i2v discount was removed upstream on 2026-06-01; only video-to-video remains cheaper). Sora 2 is flat-priced at $0.10/sec for both t2v and i2v — note it rejects human faces in reference images (use Seedance + RealFace for real people). Calls block for 30–120s while the upstream polls the job. Seedance 2.0 Fast typically returns in 60–80s; 2.0 Pro trades latency for quality.

BytePlus RealFace (Seedance 2.0 only)

For real-person character consistency across frames, pass real_face_asset_id (format ta_xxxxxxxx) on 2.0 Fast or 2.0 Pro. Asset IDs come from token360's Asset UI after H5 face verification — see blockrun's /docs/video/real-person-ip for the enrollment walkthrough. Cannot be combined with image_url (both seed the first frame — pick one). Pricing is unchanged.

curl -X POST http://localhost:8402/v1/videos/generations \
  -H "Content-Type: application/json" \
  -d '{"model":"bytedance/seedance-2.0","prompt":"the person walks through a forest","real_face_asset_id":"ta_abc123def","duration_seconds":5}'

Image Editing (img2img)

Edit existing images with /img2img:

/img2img --image ~/photo.png change the background to a starry sky
/img2img --image ./cat.jpg --mask ./mask.png remove the background

API endpoint: POST http://localhost:8402/v1/images/image2image — see full docs.

Phone & Voice Calls

Verify phone numbers and place AI-powered outbound voice calls directly from chat. Phone intelligence runs on Twilio; voice calls use Bland.ai. Payment is automatic via x402 from the wallet.

/cr-call +14155552671 "Hi, this is calling to confirm tomorrow's 3pm meeting"
/cr-call +14155552671 "Order a large pepperoni for delivery" --voice josh --max-duration 10

Calls are fire-and-forget: the request returns a call_id and poll_url immediately. The call itself runs in the cloud for up to 30 minutes. Poll GET /v1/voice/call/{call_id} (or clawrouter share/transcripts dashboard) to retrieve the transcript and recording when status is completed.

CLI for wallet-owned numbers:

clawrouter phone numbers list                              # See active numbers + expiry
clawrouter phone numbers buy US --area-code 415            # Provision a SF number
clawrouter phone numbers renew +14155551234                # Extend 30 days
clawrouter phone numbers release +14155551234              # Release
clawrouter phone lookup +14155552671                       # Carrier + line type
clawrouter phone fraud +14155552671                        # SIM-swap + fwd signals

HTTP API:

# Place a call
curl -X POST http://localhost:8402/v1/voice/call \
  -H "Content-Type: application/json" \
  -d '{"to":"+14155552671","task":"Confirm the 3pm Thursday meeting.","max_duration":5}'
# → { "call_id": "call_abc123", "poll_url": "/v1/voice/call/call_abc123", "status": "queued" }

# Poll for transcript
curl http://localhost:8402/v1/voice/call/call_abc123

LLM agents discover all eight operations as blockrun_phone_* / blockrun_voice_* tools (see /partners).

⚠️ blockrun_voice_call and /cr-call place a real outbound phone call. Server enforces an emergency-number blocklist; choose --from from wallet-owned numbers via phone numbers list.

Crypto Data (Surf)

Surf is BlockRun's unified crypto data API — 84 endpoints across 13 domains: CEX/DEX markets, on-chain SQL over 80+ ClickHouse tables (Ethereum, Base, Arbitrum, BSC, TRON, HyperEVM, Tempo), 100M+ labeled wallets, prediction markets (Polymarket + Kalshi), social/CT mindshare, news, project/DeFi metrics, token analytics, unified search, VC fund intelligence. The killer feature is ad-hoc POST /surf/onchain/sql — agents query the warehouse directly without running an indexer.

ClawRouter ships Surf as a skill, not as typed wrappers. The proxy whitelists /v1/surf/* so any call through the local proxy is paid x402 from the same wallet; the agent reads skills/surf/SKILL.md for the endpoint catalog and crafts the HTTP call. No blockrun_surf_* tool definitions to maintain; a new Surf endpoint requires zero ClawRouter release.

Usage (HTTP):

# Aggregated BTC spot price (Tier 1, $0.001)
curl 'http://localhost:8402/v1/surf/market/price?symbol=BTC'

# Bulk wallet labels over 100M+ labeled wallets (Tier 2, $0.005)
curl 'http://localhost:8402/v1/surf/wallet/labels/batch?addresses=0xabc,0xdef,0x123'

# Ad-hoc on-chain SQL (Tier 3, $0.020)
curl -X POST 'http://localhost:8402/v1/surf/onchain/sql' \
  -H 'content-type: application/json' \
  -d '{"sql":"SELECT count() FROM ethereum.transactions WHERE block_timestamp >= now() - INTERVAL 1 HOUR"}'

No Surf account, no API key — settles directly to Surf's Base treasury in USDC via the same wallet as LLM calls. Full endpoint reference: skills/surf/SKILL.md. Upstream marketplace: https://blockrun.ai/marketplace/surf.

Models & Pricing

55+ models across 9 providers, one wallet. Starting at $0.0002/request.

💡 "Cost per request" = estimated cost for a typical chat message (~500 input + 500 output tokens).

Budget Models (under $0.001/request)

Mid-Range Models ($0.001–$0.01/request)

Premium Models ($0.01+/request)

Free tier: several NVIDIA-hosted models cost nothing — /model free smart-routes across them, or pick one directly (e.g., /model nemotron-omni for vision, /model qwen3-next for 262K reasoning, /model qwen-coder for coding, /model maverick). Best value: gpt-5-nano and gemini-2.5-flash-lite deliver strong results at ~$0.0003/request.

Payment

No account. No API key. Payment IS authentication via x402.

Request → 402 (price: $0.003) → wallet signs USDC → retry → response

USDC stays in your wallet until spent — non-custodial. Price is visible in the 402 header before signing.

Dual-chain support: Pay with USDC on Base (EVM) or USDC on Solana. Both wallets are derived from a single BIP-39 mnemonic on first run.

/wallet              # Check balance and address (both chains)
/wallet export       # Export mnemonic + keys for backup
/wallet recover      # Restore wallet from mnemonic on a new machine
/wallet solana       # Switch to Solana USDC payments
/wallet base         # Switch back to Base (EVM) USDC payments
/chain solana        # Alias for /wallet solana
/stats               # View usage and savings
/stats clear         # Reset usage statistics
/exclude             # Show excluded models
/exclude add <model> # Block a model from routing (aliases work: "grok-4", "free")
/exclude remove <model> # Unblock a model
/exclude clear       # Remove all exclusions

Fund your wallet:

• Base (EVM): Send USDC on Base to your EVM address
• Solana: Send USDC on Solana to your Solana address
• Coinbase/CEX: Withdraw USDC to either network
• Credit card: Reach out to @bc1max on Telegram

Screenshots

Smart Routing in Action

Telegram Integration

Configuration

For basic usage, no configuration needed. For advanced options:

Full reference: docs/configuration.md

Model Exclusion

Block specific models from being routed to. Useful if a model doesn't follow your agent instructions or you want to control costs.

/exclude add nvidia/gpt-oss-120b   # Block the default free model
/exclude add grok-4                # Aliases work — blocks all grok-4 variants
/exclude add gpt-5.4               # Skip expensive models
/exclude                           # Show current exclusions
/exclude remove grok-4             # Unblock a model
/exclude clear                     # Remove all exclusions

Exclusions persist across restarts (~/.openclaw/blockrun/exclude-models.json). If all models in a tier are excluded, the safety net ignores the filter so routing never breaks.

Troubleshooting

When things go wrong, run the doctor:

npx @blockrun/clawrouter doctor

This collects diagnostics and sends them to Claude Sonnet for AI-powered analysis:

🩺 BlockRun Doctor v0.12.24

System
  ✓ OS: darwin arm64
  ✓ Node: v22.14.0

Wallet
  ✓ Address: 0x1234...abcd
  ✓ Balance: $12.50

Network
  ✓ BlockRun API: reachable (142ms)
  ✗ Local proxy: not running on :8402

📤 Sending to Claude Sonnet 4.6 (~$0.003)...

🤖 AI Analysis:
The local proxy isn't running. Run `openclaw gateway restart` to fix.

Use Opus for complex issues:

npx @blockrun/clawrouter doctor opus

Ask a specific question:

npx @blockrun/clawrouter doctor "why is my request failing?"
npx @blockrun/clawrouter doctor opus "深度分析我的配置"

Cost: Sonnet ~$0.003 (default) | Opus ~$0.01

Development

git clone https://github.com/BlockRunAI/ClawRouter.git
cd ClawRouter
npm install
npm run build
npm test

Support

From the BlockRun Ecosystem

More Resources

Blog

Frequently Asked Questions

What is ClawRouter?

ClawRouter is an open-source (MIT licensed) smart LLM router built for autonomous AI agents. It analyzes each request across 15 dimensions and routes to the cheapest capable model in under 1ms, entirely locally — no external API calls needed for routing decisions.

How much can ClawRouter save on LLM costs?

ClawRouter's blended average cost is $2.05 per million tokens compared to $25/M for Claude Opus, representing 92% savings. Actual savings depend on your workload — simple queries are routed to free models ($0/request), while complex tasks get premium models.

How does ClawRouter compare to OpenRouter?

ClawRouter is open source and runs locally. It uses wallet-based authentication (no API keys) and USDC per-request payments (no credit cards or subscriptions). OpenRouter requires an account, API key, and credit card. ClawRouter also features smart routing — it automatically picks the best model for each request, while OpenRouter requires manual model selection.

How does ClawRouter compare to LiteLLM?

Both are open source and run locally. But ClawRouter adds smart routing (automatic model selection), wallet-based auth, and USDC payments. LiteLLM requires you to bring your own API keys and manually choose models.

What agents does ClawRouter work with?

ClawRouter works with any tool that makes OpenAI-compatible API calls — point it at http://localhost:8402. This includes continue.dev, Cursor, VS Code extensions, ElizaOS, and custom agents. It also integrates as a plugin with OpenClaw (an AI coding agent), which enables additional features like slash commands and usage reports.

Is ClawRouter free?

ClawRouter itself is free and MIT licensed. You pay only for the LLM API calls routed through it — and several NVIDIA-hosted models (gpt-oss-120b, gpt-oss-20b, mistral-large-3-675b, qwen3.5-122b-a10b, qwen3-next-80b-a3b-instruct, llama-4-maverick, seed-oss-36b, nemotron-3-nano-omni-30b-a3b-reasoning) are completely free. Use /model free to smart-route across them, or pick any by name.