Surfaced

We built an AI agent that books local SF services by voice. Type prompt it searches Senso, verifies with Tinyfish, and Vapi calls to book. One prompt, done. Providers live on Cited.md, open to any AI.

Comment

Inspiration

Finding and booking local services is broken. You search Yelp, call a number that's wrong, leave a voicemail, and wait. We wanted one prompt to do all of it.

What it does

SF Service Agent books local services by voice. Type "massage under $90" — the agent searches a verified provider database, confirms the real phone number against the live web, then calls and books on your behalf.

How we built it

  • Senso — provider listings stored as cited, agent-readable markdown on Cited.md
  • Tinyfish — web agent verifies each provider's real phone number via Google
  • Vapi — voice AI makes the actual booking call
  • Python — orchestration layer connecting all three

Challenges

Tinyfish verification added latency. We solved it by routing directly to a provider's website URL when available, falling back to Google search only when needed.

What we learned

Agent-to-agent data pipelines need a trusted source of truth. Cited.md is that layer — providers publish once, any agent can query and pay for the data.

What's next

Expand beyond SF. Add real-time availability. Let providers monetize their listings via x402 payment rails.

Built With

  • cited.md
  • python
  • rest
  • senso
  • tinyfish
  • vapi
Share this project:

Updates