Haggler

Haggler: snap the damage, and your AI agent gets the body shops competing.

After a crash you usually get one estimate and no leverage. With Haggler, a driver uploads photos and sends their AI agent to shop the repair. The agent looks at the photos and writes a damage report. Then every body shop's own AI agent inspects the same photos independently and builds its own repair list in its own style. Our server prices each list with that shop's rate card.

One shop is real: Drive Auto Body, my dad Gio's shop. Its agent is grounded in 115 anonymized jobs mined from his real estimates. On insurance jobs it finds the hidden damage, scans and blends he historically adds (for example, he wrote a pre-repair scan on 91% of 68 front-end insurance jobs). On cash jobs it picks the cheapest sound fix (he repaired the hood instead of replacing it on 4 of 4 cash jobs).

The driver's agent questions gaps between estimates, then asks every shop for its best and final. Each shop agent has limits it can't cross alone. Past them, it asks its owner, who taps Approve, Counter or Deny in that shop's own Slack room with the photos and damage list in front of them. Any shop can win, and nothing books until the driver confirms.

Built on ZooWork managed agents (Claude Sonnet 5 vision through custom tools; the driver's agent plus one agent per shop, each bound to its identity on our server). BAND mirrors the negotiation into a shared agent room with each agent as itself. Tavily pulls real shop info (Drive's 4.9-star, 49-review rating today; warranties and certifications next). The server enforces every limit: single-use approvals tied to one exact offer, range-checked counters, rate limits so nobody drains credits, and honest insurance wording. o