BlockRun is the payment rail for AI — a service marketplace where AI agents discover, route, and pay for models, data APIs, and secure runtimes using USDC via x402.

How does BlockRun work?

BlockRun uses the x402 protocol for HTTP-native payments. AI agents send API requests, receive a 402 Payment Required response with the price, sign a USDC payment locally, and retry with the payment header.

How much does BlockRun cost?

BlockRun charges provider cost plus 5% margin. There are no subscriptions or minimum commitments — you pay only for what you use.

“Agent friendly LLM optimizations.”— Jeremy Allaire, Circle

YOPO.

You Only Pay for the Outcome.

Plug your agent into 55+ models, live data APIs, and secure code runtime. Pay per call in USDC via x402. No API keys, no accounts, no subscriptions.

Install ClawRouter Try Franklin 6K ★

🦞ClawRouter won Agentic Commerce · World's first AI hackathon by USDC

1.1M+ monthly calls·50+ countries·55+ models·

live

$npm i -g @blockrun/clawrouter

works with your existing wallet — no API keys

Docs ↗

curl -X POST https://blockrun.ai/api/v1/chat/completions

Routes to

ClawRouter · open source

Cut AI costs
by up to 78%.

A smart router for LLM calls. Routes to the cheapest capable model, compresses prompts, and caches responses. Drop-in OpenAI-compatible.

Install from npm Read docs ↗

MIT licensed · built on Base

clawrouter.config.tssaved ✓

// route by task type → cost ceiling
export default defineRouter({
  fallback: "deepseek-v3.2",
  rules: [
    { task: "code",      model: "claude-sonnet-4.6",  max: 0.02 },
    { task: "reasoning", model: "gpt-5.4",            max: 0.03 },
    { task: "summarize", model: "gemini-2.5-flash",   max: 0.005 },
  ],
  compress: true, cache: true,
});

Smart RoutingRoutes to cheapest capable model

7-Layer Compression15–40% token reduction

Response Caching~200ms latency savings

Native OpenClawSessions, plugins, memory

FLAGSHIP AGENTfranklin.runYOPO — You Only Pay Outcome

The AI agent with a wallet.

Other agents write code. Franklin writes code and spends money to get things done. It holds your USDC, picks the best model per task, buys trading data, generates images, searches the web — and decides what's worth paying for. You set the budget. It runs it.

55+ models: Claude, GPT-5, Gemini, DeepSeek, Grok, and more
Wallet-native: USDC on Base + Solana · budget caps enforced on-chain
Smart Router: Auto-selects best model per task — up to 89% savings vs Opus
Self-evolution: Learns your preferences locally across sessions
Migration tool: `franklin migrate` imports Claude Code sessions

Visit franklin.run View on GitHub ↗

franklin · ~/workYOPO · wallet live

wallet

0x7a…3f2c

USDC

128.4032

session · 0

− 0.0000

❯$ franklin --trust

live demo

$curl -fsSL https://franklin.run/install | shnode 20+ · macos · linux · wsl

“

Best part is that Agents and Devs can use this on usage basis using USDC on x402. Agent friendly LLM optimizations.

Jeremy Allaire

CEO, Circle · Co-founder of USDC

Read thread ↗

MCP Server

One install.
8 agent tools.

Drop BlockRun into Claude Code, Cursor, or any MCP client. Your agent gets models, live data, media generation, and a secure runtime — all paid with USDC.

$claude mcp add blockrun --transport http https://mcp.blockrun.ai/mcpstdio · http

blockrun_chat

44 AI models

blockrun_exa

Neural search

blockrun_search

Live web/X/news

blockrun_markets

Prediction markets

blockrun_image

DALL·E, Nano Banana, Grok Imagine

blockrun_video