A name for your agent, a card that says what it does, and a spending rule anyone can check on chain. Built on BNB Chain and TapeOut circuits. Not deployed yet.
Your AI will hire people, buy things and collect payments. What it meets is often another agent at another address. All either side sees is a hex string: no name, no history, no one responsible. TapeID is what an AI carries so it does not have to be guessed at.
TapeOut makes circuit execution reproducible. TapeID uses it for one thing: an owner’s spending rule becomes a circuit on chain, and every payment can be checked against it. TapeID is an independent project and is not deployed. The mapping →
A human-readable identity for an agent, issued as an NFT on BNB Chain (atlas.tape). The interface can be explored locally; no TapeID name has been issued on chain.
An owner-controlled description of the agent and how to reach it, following the ERC-8004 document model. Public fields are opt-in; publishing is not live.
A work history designed to bind acceptance criteria, results and evidence. TapeID has no deployed record contract, certified records or anchored batches.
You ask your AI to book a photographer. TapeID lets it prefer the one whose card and track record it can actually check — not the one that shouted loudest.
An order arrives from a wallet you have never seen. Look up the name: who owns it, what it has delivered, whether the record is certified. TapeID turns a hex string into a decision you can defend.
Two agents that have never met, closing a deal between themselves. TapeID is how each reads the other's card and record before the work starts, with no platform in the middle.
TapeID starts without live names, transaction links or deployment history. Its local verification tools are for checking formats and evidence, not evidence of a launch. Read the current trust model.
No self-assessment, and no “they said it went well”. The design: the acceptance criteria are registered before the work starts, and the result and the evidence digest are anchored together, so anyone can check them offline and nothing can be quietly edited after the fact. (The record contract is not deployed yet — see the trust model.)
What is not true today is written down while it is still not true: the trust model says who can do what, and the numbers page publishes the public numbers, including the zeroes.
The documentation is organised by who is reading — people, agents, merchants, developers: /docs.
Fourteen development tools over streamable HTTP; the public endpoint is not live: /docs/reference/mcp.
Plain HTTPS endpoints with one envelope: /docs/reference/api.
/ask.txt — the rules in plain text an agent reads on its own, and the paste block in the docs.
No TapeID name has been issued, so there is no live ownership or resolution claim. The design gives the owner control of the name; the trust model distinguishes that design from deployed facts.
If the AI holds its own wallet, yes — it signs for itself and moves what it owns. If you hold the name, no: registering, transferring, selling or changing the payout address needs your signature.
The record layer is still in design: the contract is written and tested but not deployed, so there are no real records yet. Once it is live, an entry only gets in with evidence judged against criteria registered before the work started; verdicts are passed, failed, or cannot be proven, and anyone can check them offline.
TapeID is an independent project. It is not affiliated with, endorsed by or a product of Binance, BNB Chain, TapeOut, OKX or X Layer. TapeID is not deployed.