# KaspaSeal

> Tamper-proof audit trails for AI agents: every action is declared before it
> runs (INTENT) and sealed after (RESULT), anchored as a 69-byte record on the
> Kaspa BlockDAG — verifiable by anyone, no account, no trust in KaspaSeal's
> servers. A live, install-free demo hands agents a disposable budgeted
> wallet on the TN10 test network.

Important: only the SHA-256 of a payload is written on-chain — payload
contents never touch the chain. Hosted-demo nuance: the demo backend receives
your payload server-side to anchor it; self-host the SDK if contents must
never leave your process. Proofs show what an agent declared and when — pair
with tool-side receipts when execution guarantees are needed.

## Docs

- [Agent instructions (complete API spec, PoW algorithm, lifecycle, errors)](https://kaspaseal.dev/agents.md): everything an autonomous agent needs to use the live demo API at https://demo.kaspaseal.dev
- [Universal verify page](https://verify.kaspaseal.dev): open-source, checks any proof txid against the public chain
- [Demo API status](https://demo.kaspaseal.dev/healthz): live health + capacity

## Quick facts

- Demo sessions: 50 anchors, 30 minutes, ≥2 s between anchors, payloads ≤ 4 KB
- One proof-of-work (SHA-256, 20 leading zero bits) per session; ~1 s to solve
- Action types: INTENT, COMMIT, RESULT, VERIFY, DISPUTE; chains form via parent_txid
- Network: Kaspa testnet-10 (real transactions, worthless tokens); mainnet on the roadmap