The problem it solves
AI agents need to pay for things on their own. But you can’t safely hand an agent a wallet’s private key — one bad instruction and it could be drained. Agent Wallets give the agent a constrained way to spend, so a leaked key can only do what you allowed.How it works
- Create an agent — give it a name, a daily spending limit, and an allowlist of addresses it can pay.
- Get an API key — shown once. The agent uses this as a Bearer token to spend.
- Fund the wallet — send USDC to the agent’s wallet address.
- The agent spends — it calls the spend endpoint with its key. Every transaction is checked against your rules before money moves.
Guardrails
If a spend would breach the allowlist or daily limit, it’s blocked and logged — the transfer never happens.
Spending from an agent wallet
Autonomous x402 payments
This is where Agent Wallets and Conduit’s x402 facilitator come together. An agent can autonomously discover a paid API, pay for it from its own wallet — within its guardrails — and receive the data back, with no human in the loop. Conduit handles the entire loop in a single call:{"listingId":"<marketplace-listing-id>"} as the body instead of a URL.
Behind that one request, Conduit:
- Authenticates the agent with its API key
- Reads the target endpoint’s x402 paywall (price + payment address)
- Pays that price from the agent’s wallet — enforcing the allowlist and daily limit
- Re-requests the endpoint with the on-chain payment proof
- Returns the data plus a receipt
The guardrails still apply. For an agent to pay an x402 endpoint, that endpoint’s payment address must be in the agent’s allowlist — so even autonomous API payments stay within the rules you set.

