Agentic Finance Graph
Connect

Home/Data/Self-check

Self-check · free · no sign-in

Check your own agent: everything we hold about it, in one answer

Give the token id or the wallet of an agent you run. You get its level and record, the Agent Pass, the Watch's reading of its newest payment, the token approvals its wallet has given, its detections in the last 30 days and, if it is funded, what a stolen key would control. At the top: the items worth a look, each with its source. Facts with their measurement times, never a verdict.

Agentic Finance Graph · Published · results load live from /api/self-check

Base ERC-8004 registrations. An agent appears here from its first counted payment; a registration that has never paid anyone has no record to check yet.

What is checked

  • Record and level. L7 seen paying, L8 still funded, L9 repeat counterparties; its last counted payment and any open incident. rank.l7plus.v1
  • The Agent Pass. Eight checks at one moment, valid three hours: what held, what failed, what could not be checked. Signed with Ed25519 when it holds. agent.pass.v2
  • The Watch. Its newest payment scored against its own earlier payments: none, notable or major, with the reasons that fired. behaviour.out_of_character.v2
  • Token approvals. What its wallet let others spend in the last 30 days: unlimited approvals still standing with the money behind them, and any approval to a plain account, the shape approval drainers use.
  • Detections. Drain shapes, look-alike losses, delegate changes and the other detector kinds on this agent in 30 days. How detections work
  • Blast radius. For funded agents (L8 and above): what a stolen key for its wallets controlled at one block, and whether the wallet runs under an EIP-7702 delegate. agent.blast_radius.v1

Let your agent run it

The same check answers agents directly, free and without a key.

A2A (JSON-RPC message/send at https://agenticfinancegraph.com/api/a2a): send the text self-check agent 95875, with your own id.

MCP (https://agenticfinancegraph.com/mcp): the tool self_check with {"agent": "95875"}.

HTTP: GET /api/self-check?agent=95875 returns the JSON this page shows.

What a clean result means

Nothing in our data is worth flagging at the measurement times shown. It does not mean the agent is safe. We cannot see who holds the key, what the agent is for, payments on chains other than Base, approvals older than 30 days, or anything after the times on each part.

If something is worth a look: an approval you did not mean to give can be revoked from the wallet that gave it; a detection links to its evidence; and with a Desk session you can plant canaries that tell you when a key or a tool is misused. Before your agent pays another, run the Gate.