$ GET /pricing

Pay per call.
Pay per report.

Pay in compute with a proof-of-work puzzle, or settle micro-amounts per call - USDC on Base, Solana, Polygon, Arbitrum, Monad, Celo, Avalanche, Sei, Optimism, Stellar, or Algorand - or USDG on Robinhood Chain. No signup, no minimum; card welcome at /reports and /credits, and monitors are the one subscription ($5, cancel anytime). The wallet is the identity.

FREE · PROOF-OF-WORK
$0.00 / call

Solve a short sha256 puzzle - a few seconds of CPU - instead of paying. No wallet at all.

✓ 166 pure-CPU tools
✓ No wallet, no funds, no account
✓ Hosted MCP connector runs these free
· Rate-limited; browser/GPU tools excluded
x402
PAID · USDC ON BASE
$0.001 / call & up

An x402 client signs USDC from the agent's own wallet and retries. Settles on Base in seconds.

✓ All 603 tools, including browser & memory
✓ Flat per-call price - pay exactly what you use
✓ Non-custodial on this rail - you sign, we hold no balance
✓ Spend caps refuse a runaway model before paying
// from-price by category

What each call costs.

Most tools - text, math, encoding, time, validation, convert$0.001
Agent memory - write, recall, grant, auditfrom $0.001
Payments & x402 - decode, verify, settlefrom $0.002
Article extract - clean markdown out$0.010
Headless browser - render & screenshot (real Chromium)$0.02
live machine-readable prices: GET /api/pricing · GET /openapi.json
// POST /v1/*

The /v1 LLM gateway.

Point any OpenAI SDK at base_url https://agent402.tools/v1/metered and every request is quoted from its own body before payment, then settled at what the call actually used, under that quote. Same wallet-is-the-identity model as every other tool, or a prepaid credits key as the API key. The flat tiers below stay for callers that want one fixed price; omit the model on the auto tier and the gateway picks one for the prompt.

/v1/metered/chat/completions - quoted per request, settles actual usage under the quotefrom $0.001
/v1/embeddings - default-on cache, free repeat$0.002
/v1/nano/chat/completions - high-frequency agent loops$0.003
/v1/auto/chat/completions - model optional, eval-ranked routing$0.01
/v1/chat/completions - budget/mid models$0.02
/v1/images/generations - one image per call$0.08
/v1/audio/speech - OpenAI TTS wire, mp3/pcm bytes out$0.06
/v1/pro/chat/completions - gpt-4o, claude sonnet, gemini pro$0.10
/v1/premium/chat/completions - gpt-5, gpt-6 astra, o3, claude opus, claude fable 5.1$0.50
full wire format: /docs
WHY IT'S CHEAP

Beat the token math.

Writing, testing and debugging a CSV parser or cron calculator mid-task burns thousands of tokens - easily 10–100× the price of a tested $0.001 call. Reimplementation is the expensive path.

build it yourself~5,000 tokens
call the tested endpoint$0.001
MAKING A LOT OF CALLS

Stop signing every call.

Per-call pricing is the point, but a signature per call is not. Paying exact over x402 means one authorization signed and one settlement waited on for every request, which is fine for ten calls and the wrong shape for ten thousand. Two ways to pay once and then just call, both live today, both still per-request priced and still debited only on a 200.

Prepaid credits · no wallet at all

Buy a pack by card once, then send Authorization: Bearer a402_… on any paid route. No wallet, no signature, no gas, no subscription. The list price is held before the call and debited only when it returns 200; on the metered gateway you are debited what the call actually used, not the quote.

/credits · $20 / $50 / $100
Keep the wallet · approve once

A one-time Permit2 approval for USDC on Base turns every quote into a ceiling rather than a charge: the gateway settles the actual usage under it, per request, with no fixed tier to overpay into. Non-custodial on this rail, and the approval is yours to revoke.

npx agent402-openclaw permit2-approve

Start on the free tier. No wallet.

QUICKSTART → Browse tools