toolkit.fun / AgentPay
Toolkit capability / agent execution

Give an agent a budget, then let it pay for the work it uses.

AgentPay is Toolkit's control plane for model discovery, bounded agent spending, pay-per-use execution, and recoverable usage receipts. It sits alongside Toolkit tools, workflows, REST APIs, and MCP servers.

The payment rail is part of the capability contract.

An agent should know what it may spend, what it actually used, and whether the result can be recovered. Toolkit keeps those decisions attached to the same model request and usage receipt.

01Owner policy bounds the request before execution.
02Provider usage is recorded with input and output tokens.
03Unknown outcomes reconcile without silently paying twice.

Choose how the agent is funded.

Payment methods share one policy and receipt model. Toolkit makes the funding source visible before a call runs.

Live · account or team ledger

Toolkit Credits

Use your Toolkit Cloud model allowance. AgentPay reserves, meters, settles, and reconciles against the existing credit ledger.

Live-capable · Monad testnet

x402 Wallet

Sign a bounded wallet authorization and settle actual usage with x402 upto and Permit2 on Monad Testnet when readiness is green.

Preview · adapter in progress

Provider Direct

Bring a credential reference in a future adapter. The current public AgentPay route does not accept raw provider keys or claim platform OpenRouter credentials as BYOK.

One flow for people and agents.

The same capability can be discovered through the web, REST, MCP, or an SDK integration.

01 / discoverCompare machine-fulfillable model offers.
02 / authorizeApply policy, funding source, and spending cap.
03 / executeRun the provider and measure actual tokens.
04 / reconcileSettle or recover without rebroadcasting unknown work.
05 / continueReturn a durable result and public proof where available.
Current release boundary. The public Hackathon Demo is a deterministic replay and remains available at /agentpay. Live AgentPay currently targets Monad Testnet and requires explicit readiness, provider credentials, a funded wallet, and a configured facilitator. Toolkit never treats a recorded replay as a live purchase.