Inspiration
During NandaHack, we explored how AI agents on NANDA Town need a reliable way to prove what happened between them. When two agents negotiate and reach a deal — who verifies it? How does Agent B know Agent A kept their promise? Town Notary solves this: it timestamps agreements, issues cryptographically signed receipts, and lets agents build reputation scores based on fulfilled vs disputed interactions. It's a trust layer for the agent economy.
What it does
Town Notary is a hosted API service with four core features:
- Notarize — Agents submit data (agreements, statements, logs) and get back a unique ID, SHA-256 hash, timestamp, and HMAC signature proving the record was created by Town Notary.
- Verify — Any agent checks if a record exists and retrieves its full contents.
- Report — After a deal executes, agents report whether it was fulfilled or disputed.
- Reputation — Query any agent's trust score (0.0–1.0) based on their fulfillment history.
How we built it
- FastAPI for the API layer — lightweight, fast, great DX
- SQLite for persistence (ephemeral on Vercel, persistent locally)
- HMAC-SHA256 for signed receipts — agents can verify offline that a record came from the notary
- Vercel for hosting — serverless Python deployment
- Dark-theme UI dashboard at the root URL for human interaction with all endpoints
Challenges we faced
- SQLite on Vercel's read-only filesystem — fixed by using
/tmp/for the database - Vercel's Python serverless runtime with ASGI/FastAPI — wrapped with Mangum for compatibility
- Making the API simple enough that a stock AI agent can use it from just the SKILL.md file ## What's next for Town Notary
Log in or sign up for Devpost to join the conversation.