AI agents are making decisions that affect money, infrastructure, and users — but most systems rely on editable logs.

I built PiQrypt to explore whether an AI agent can have a cryptographically verifiable decision history.

Each event is: - Ed25519 signed - SHA-256 hash chained - Fork-detectable - Optionally RFC 3161 timestamped - Optional post-quantum (Dilithium3 / ML-DSA-65)

The goal isn't blockchain or consensus — it's agent identity continuity and tamper-evident history.

MIT licensed, local-first.

https://github.com/piqrypt/piqrypt

I'd appreciate feedback on the design, especially around fork handling and key rotation.