Skip to content

Repository files navigation

pantessa

Renamed from yeetful. Yeetful is now Pantessa. npm i pantessa && npm rm yeetful, swap the import specifier, and you're done — every renamed export keeps its old name as a deprecated alias. The yeetful package lives on as a thin re-export and will not get further fixes; upgrading also gets you the hosted defaults on the current domain and an embed origin check that survives the redirect.

Spend-controlled x402 for AI agents. Give an agent an expense account — an allowlist of endpoints plus per-call / per-day budgets — and let it pay any x402 service with no API keys. Enforcement is local and instant; every call emits a receipt. Built for Pantessa, MIT-licensed, works anywhere TypeScript does.

npm install pantessa viem

Agent expense account

Wrap your agent's calls in one grant-aware pay(). It refuses anything off the allowlist or over budget before signing a payment — your guardrail against runaway loops, bugs, and prompt-injected tool calls.

import { pantessa, GrantError } from 'pantessa/agent'
import { createWalletClient, http } from 'viem'
import { base } from 'viem/chains'
import { privateKeyToAccount } from 'viem/accounts'

const wallet = createWalletClient({
  account: privateKeyToAccount(process.env.PRIVATE_KEY as `0x${string}`),
  chain: base,
  transport: http(),
})

const pay = pantessa({
  wallet,
  grant: {
    allow: ['tripadvisor.x402.paysponge.com', 'anthropic.yeetful.com'],
    perCallUsd: 0.05,
    perDayUsd: 2,
    expiresAt: '2026-12-31',
  },
  onReceipt: (r) => console.log(r.host, `$${r.amountUsd}`, r.txHash ?? r.note),
})

try {
  const res = await pay('https://tripadvisor.x402.paysponge.com/api/v1/location/search?searchQuery=tokyo')
  console.log(await res.json())
  console.log(`spent today: $${pay.spentTodayUsd()} / left: $${pay.remainingTodayUsd()}`)
} catch (e) {
  // GrantError.code: NOT_ALLOWED | OVER_PER_CALL | BUDGET_EXCEEDED | EXPIRED | REVOKED
  //   | OVER_AGENT_BUDGET | OVER_ORG_BUDGET | AGENT_PAUSED | ACCOUNT_FROZEN
  if (e instanceof GrantError) console.error(`blocked: ${e.code}`)
}

One grant authorizes many endpoints (the allowlist). Use onReceipt to stream the audit trail to your dashboard or the Pantessa control plane.

Hosted-ledger sync

Mirror a grant you created at pantessa.com and pass an API key (minted on the dashboard) — every receipt then syncs to your hosted ledger, so budgets and the audit feed include this agent's calls:

const pay = pantessa({
  wallet,
  grant: { id: 'your-grant-id', allow: [...], perCallUsd: 0.05, perDayUsd: 2 },
  apiKey: process.env.PANTESSA_API_KEY, // yf_…
})
// …
await pay.flushLedger() // before a short-lived script exits

Sync is best-effort and never blocks or fails a payment; denials are synced too (ok: false with the violation code).

ledgerUrl must be the canonical origin (currently https://www.pantessa.com): fetch silently drops the Authorization header when it follows a cross-origin redirect such as apex → www. If sync or the policy fetch fails after a redirect, the onEvent log names the origin to use.

Per-key agent budgets

On pantessa.com an agent is an API key — the dashboard's Agents tab gives each key a per-day USD budget and a spent-today meter. When you pass apiKey, the SDK fetches the key's policy (GET /api/agent/policy) before the first payment and refuses to pay once the key is over budget, or when a call's quoted price would exceed what's left today:

const pay = pantessa({ wallet, grant: { id: 'your-grant-id', ... }, apiKey: process.env.PANTESSA_API_KEY })

console.log(pay.agentBudget()) // { keyId, label, perDayUsd, spentTodayUsd, remainingTodayUsd, overBudget }
// over budget → pay() throws GrantError('OVER_AGENT_BUDGET') and syncs the
// denial receipt, so the refusal shows up in the dashboard audit trail.

Budgets are advisory at the rails — the agent pays from its own wallet, so this local refusal is the enforcement point. The snapshot stays fresh opportunistically: receipt-sync responses echo the updated budget, flushLedger() re-fetches the policy (picking up mid-run dashboard edits), and settled-but-unsynced spend is counted locally in between. If the policy can't be fetched at all, payments proceed under the grant alone.

Local vs. hard enforcement. This SDK enforces the grant in-process — ideal for governing your own agents (runaway loops, bugs, injected tool calls). For adversarial guarantees, back the grant with an on-chain Coinbase Spend Permission so the wallet contract caps spend regardless of the SDK.

Org budgets & remote pause (0.5)

If the key belongs to an organization on pantessa.com, the same apiKey flow adds two more controls — fetched from the policy, refreshed on every sync echo, and enforced locally just like the per-key budget:

  • Two-level budget. The org has a daily USD cap above each key's own budget — summed across all the org's agents. A call that would breach it throws GrantError('OVER_ORG_BUDGET'). Over either level stops the payment.
  • Remote kill switch. An admin can freeze a single agent (AGENT_PAUSED) or the whole expense account (ACCOUNT_FROZEN) from the dashboard. The SDK halts all payments while frozen — a hard stop above any budget arithmetic — and resumes automatically on the next policy refresh once unfrozen.
const pay = pantessa({ wallet, grant: { id: 'your-org-grant-id', ... }, apiKey: process.env.PANTESSA_API_KEY })

pay.orgBudget() // { id, name, perDayUsd, spentTodayUsd, overBudget } | null (null for personal keys)
pay.status()    // { halted, haltReason: 'AGENT_PAUSED' | 'ACCOUNT_FROZEN' | null }

// org over its cap   → GrantError('OVER_ORG_BUDGET')
// agent/account paused → GrantError('AGENT_PAUSED' | 'ACCOUNT_FROZEN'), before any network call

Same honesty as budgets: pause is advisory at the rails for SDK agents paying their own wallet (this local refusal is the enforcement); the chats Pantessa itself executes are hard-stopped server-side, and on-chain hard stops arrive with Spend Permissions.


Low-level x402 primitives

The agent wrapper is built on a full x402 toolkit you can use directly:

// Server — gate a route for 1¢ USDC
import { withPayment } from 'pantessa/next'

export const GET = withPayment(
  { price: '0.01', recipient: '0xYourAddress', network: 'base' },
  async () => Response.json({ secret: 'gm' })
)
// Client — auto-pay when a server returns 402 (no grant enforcement)
import { createPaymentClient } from 'pantessa/client'

const pay = createPaymentClient({ wallet })
const res = await pay('https://api.example.com/premium')
console.log(await res.json()) // → { secret: 'gm' }

Why x402?

x402 is a reborn HTTP 402 Payment Required — a protocol where servers quote a price, clients sign a stablecoin authorization, and a facilitator settles on-chain. No accounts, no Stripe dashboards, no webhook retries. Works on EVM chains today (USDC on Base, Optimism, Arbitrum, Polygon, Ethereum).

You get:

  • Per-request pricing for any API — LLM calls, data feeds, premium endpoints, MCP tools.
  • One-sentence paywalls for agents: an LLM with a wallet can now pay for what it uses.
  • Instant settlement on L2 — no chargebacks, no holds, no 30-day payout delay.

Install

npm install pantessa viem
# or
pnpm add pantessa viem
# or
yarn add pantessa viem

viem is a peer dependency so the SDK stays light and stays in sync with whatever viem version your app already uses.


Quickstart

Server: gate a route

Next.js (App Router)

// app/api/premium/route.ts
import { withPayment } from 'pantessa/next'

export const GET = withPayment(
  {
    price: '0.01',                        // USD
    recipient: '0xYourWalletAddress',     // gets paid
    network: 'base',                      // or ['base', 'optimism']
    description: 'Premium GM endpoint',
  },
  async (req) => {
    return Response.json({ message: 'gm, thanks for the cent' })
  }
)

Express

import express from 'express'
import { paymentRequired } from 'pantessa/express'

const app = express()

app.get(
  '/premium',
  paymentRequired({
    price: '0.01',
    recipient: '0xYourWalletAddress',
    network: 'base',
  }),
  (req, res) => {
    res.json({ message: 'gm', payer: req.x402?.payer })
  }
)

app.listen(3000)

Anywhere else (Hono, Bun, Cloudflare Workers, raw Node)

Use the runtime-agnostic gate() helper. Give it a standard Request, get back either a 402 Response or a settle() handle.

import { gate } from 'pantessa/server'

export default {
  async fetch(request: Request) {
    const result = await gate(request, {
      price: '0.01',
      recipient: '0xYourWalletAddress',
      network: 'base',
    })

    if (result.type === 'paymentRequired') return result.response

    // …do the paid work…
    const body = Response.json({ message: 'gm' })

    const { header } = await result.settle()
    body.headers.set('X-PAYMENT-RESPONSE', header)
    return body
  },
}

Server: track earnings on your dashboard

Claimed your MCP on pantessa.com? Report each paid call so your earnings — total, last 30 days, calls served, paying agents — show up on your dashboard. reportUsage() is fire-and-forget: it never throws and never blocks, so call it after settle() and don't await it on the hot path (on serverless, hand it to ctx.waitUntil(...)).

import { gate, reportUsage } from 'pantessa/server'

const { payer, settle } = /* …from gate() … */
const { header, result } = await settle()

// non-blocking — do NOT await on the request's critical path
reportUsage({
  apiKey: process.env.PANTESSA_API_KEY!, // a yf_… key from dashboard/keys
  mcp: 'your-server-slug',              // your slug on pantessa.com/servers/<slug>
  amountUsd: 0.01,
  payer,
  tool: 'list_proposals',
  network: 'base',
})

Full walk-through: pantessa.com/docs/earn.

Client: auto-pay

import { createPaymentClient } from 'pantessa/client'

const pay = createPaymentClient({
  wallet,                           // any viem WalletClient
  maxAmountAtomic: 1_000_000n,      // cap: 1 USDC per call
  allowedNetworks: ['base'],        // only pay on Base
  onPaymentRequired: async (req) => {
    console.log(`Pay ${req.maxAmountRequired} to ${req.payTo}?`)
    return true                     // return false to cancel
  },
})

// Use exactly like fetch.
const res = await pay('https://api.example.com/premium')

Configuration

RouteGateOptions — server

Option Type Default Notes
price string | number required USD amount, e.g. '0.01'. Converted to USDC atomic units.
recipient Address required Address that receives the payment.
network X402Network | X402Network[] 'base' Networks you'll accept. Multi-chain = multi-item discovery.
asset Address USDC for network Override to use a different ERC-20.
description string Shown to the paying client.
maxTimeoutSeconds number 600 Validity window of the signed authorization.
facilitator FacilitatorConfig | false hosted facilitator Pass false to skip on-chain settlement (testing only).

Supported networks: base, base-sepolia, ethereum, optimism, arbitrum, polygon.

ClientOptions — client

Option Type Notes
wallet WalletClient Any viem wallet capable of signing EIP-712 typed data.
maxAmountAtomic bigint Reject requirements above this cap — safety belt.
allowedNetworks X402Network[] Only pay on these networks.
onPaymentRequired (req) => boolean | Promise<boolean> Approval hook; return false to cancel.
fetch typeof fetch Override the underlying fetch (e.g. for timeouts).

How it works

  1. Client requests a paid resource normally.
  2. Server responds with 402 Payment Required and a JSON body listing acceptable requirements (network, asset, amount, recipient).
  3. Client picks the cheapest requirement, signs an EIP-3009 TransferWithAuthorization with the user's wallet, and retries the request with an X-PAYMENT header (base64 JSON).
  4. Server hands the signed payload to a facilitator which verifys the signature and settles the transfer on-chain.
  5. Server runs the handler and returns the response with an X-PAYMENT-RESPONSE header containing the transaction hash.

The signing is gasless for the payer — the facilitator broadcasts the transfer and picks up gas.


Facilitators

By default the SDK uses the hosted facilitator at https://facilitator.yeetful.com. Override it anywhere you configure the server:

withPayment(
  {
    price: '0.01',
    recipient: '0xYourAddress',
    facilitator: {
      url: 'https://your-facilitator.example.com',
      authHeader: 'Bearer your-token',
    },
  },
  handler,
)

Pass facilitator: false to skip verification and settlement entirely — only useful for local testing.


Advanced

Accept multiple networks

withPayment(
  {
    price: '0.01',
    recipient: '0xYourAddress',
    network: ['base', 'optimism', 'arbitrum'],
  },
  handler,
)

Clients automatically pick the cheapest network they're configured to use.

Sign a payment manually

import { signPayment } from 'pantessa/client'

const payment = await signPayment(wallet, {
  scheme: 'exact',
  network: 'base',
  asset: '0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913', // USDC
  maxAmountRequired: '10000', // 0.01 USDC
  payTo: '0xRecipient',
})

Use with AI agents / MCP tools

x402 is a natural fit for agent tooling — drop withPayment in front of any MCP tool endpoint and agents with wallets can pay per-call. This SDK is what powers paid tools on Pantessa.


Embed the chat

pantessa/embed drops the Pantessa chat into any webpage as an iframe — zero dependencies, framework-agnostic, browser-only (it never imports viem or the payment stack). Scope it to up to 4 MCPs with mcps, or float it as a bottom-right bubble with mode: 'bubble'.

Plain script tag:

<div id="pantessa-chat" style="height: 560px"></div>
<script type="module">
  import { mountPantessaChat } from 'https://esm.sh/pantessa/embed'

  const chat = mountPantessaChat({
    container: '#pantessa-chat',        // element or selector (inline mode)
    mcps: ['uniswap-free'],            // scope the chat to these MCPs (≤4)
    wallet: 'auto',                    // bridge window.ethereum into the chat (the default)
    theme: 'dark',
    onEvent: (name, data) => console.log('pantessa event', name, data),
  })
  // later: chat.sendPrompt('…') · chat.destroy()
</script>

The host-wallet bridge (wallet, new in 0.9): the SDK relays the host page's EIP-1193 provider into the iframe over postMessage, so the embedded chat can request accounts, read balances, and pop the user's own wallet for signatures — no separate connect flow inside the embed. The host page already holds that provider, so the bridge grants the embed the same dapp-level access the host has, nothing more: every signature/transaction still opens the USER's wallet UI for approval, relayed reads are restricted to a strict method allowlist, and no private key material ever crosses the frame. 'auto' (default) uses window.ethereum when present; pass a provider (e.g. from wagmi) or false to turn the bridge off. setAddress remains for context-only hosts that just want to tell the chat which address to talk about without wiring a wallet.

React (e.g. a CoW Swap fork), mounting in a useEffect and syncing the connected account:

import { useEffect, useRef } from 'react'
import { mountPantessaChat, type PantessaChatHandle } from 'pantessa/embed'

function PantessaChat({ address }: { address?: string }) {
  const ref = useRef<HTMLDivElement>(null)
  const chat = useRef<PantessaChatHandle | null>(null)

  useEffect(() => {
    chat.current = mountPantessaChat({
      container: ref.current!,
      mcps: ['cow-swap'],
      address,               // initial context goes in the URL
      theme: 'dark',
    })
    return () => chat.current?.destroy()
  }, []) // mount once

  useEffect(() => {
    chat.current?.setAddress(address ?? null) // queued until the embed is ready
  }, [address])

  return <div ref={ref} style={{ height: 560 }} />
}

mountPantessaChat(options) returns a handle: { iframe, setAddress, setTheme, sendPrompt, open, close, destroy }. open/close drive the bubble panel (no-ops inline); sendPrompt(text) injects a prompt as the user's message — wire it to host CTAs like an "ask about this order" button (pass { submit: false } to only prefill the input); destroy removes all DOM nodes and listeners. Security: the parent only accepts postMessage events from the embed origin with source: 'yeetful-embed', and always posts back with an explicit targetOrigin (never '*').


API reference

pantessa/server

  • gate(request, options) — runtime-agnostic. Returns { type: 'paymentRequired', response } or { type: 'ok', payer, settle }.
  • reportUsage(options) — fire-and-forget earn-side receipt to your Pantessa dashboard. Never throws; resolves true on a 2xx.
  • Facilitator — thin wrapper around verify/settle HTTP endpoints.
  • DEFAULT_FACILITATOR_URL — the hosted facilitator URL.
  • DEFAULT_RECEIPTS_URL — the hosted earn-side ingestion URL.

pantessa/next

  • withPayment(options, handler) — wraps a Next.js route handler.

pantessa/express

  • paymentRequired(options) — returns an Express RequestHandler. Sets req.x402.payer after successful verification.

pantessa/client

  • createPaymentClient(options) — returns a fetch-compatible function that handles 402s automatically.
  • signPayment(wallet, requirement) — sign a payment payload by hand.
  • PaymentError — thrown when the client declines to pay.

pantessa/embed

  • mountPantessaChat(options) — mounts the Pantessa chat iframe (inline or bubble); returns a PantessaChatHandle (setAddress / setTheme / sendPrompt / open / close / destroy). Browser-only, zero deps. options.wallet: 'auto' | Eip1193Provider | false (default 'auto') bridges the host page's wallet provider into the chat — allowlisted EIP-1193 methods are relayed over postMessage; signatures/txs always pop the user's own wallet UI.

Helpers

  • usdcAddress(network) — canonical USDC contract for a supported network.
  • usdToAtomic(amount, decimals?) — safe USD → atomic-units conversion.
  • encodePayment / decodePayment — base64 JSON codec for headers.

Development

npm install
npm run build     # bundles ESM + CJS + d.ts via tsup
npm run typecheck
npm test

To publish:

npm run build
npm publish

License

MIT © Pantessa

About

Payments for agents

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages