Coordinate multi-agent marketplaces with shared action policy

In a multi-agent marketplace, one agent may buy work from another — for example a travel agent purchasing a weather micro-forecast. Treat each sensitive request as an ActionIntent: evaluate against shared marketplace policy, write a privacy-safe audit trail, then allow or deny. Do not hold funds on behalf of agents for this.

Apply shared policy before side effects

1. The buying agent drafts an intent (action type, resource, optional cost).
2. Coordinators submit the intent to Limetry for server-authoritative evaluation.
3. Velocity, allowlists, and pending-intent limits apply across collaborating agents.
4. An allow decision (optionally with a verified receipt) lets middleware run the tool; settlement stays on your rails.

Policy and privacy-safe telemetry come first. Payment rails remain optional and separate.

Inspect the TypeScript coordinator in the multi-agent marketplace example.