Curbi / Developer platform
A record you can check.
A failure you can replay.
Portable agent action evidence and repeatable integration repairs, exposed through useful developer tools. Start with a working example, then inspect the contract underneath.
Choose the task.
| I need to… | Start here | Current scope |
|---|---|---|
| Check what an agent says | Policy Check · Check API | Configured rules and supplied records; a pass is not authority or general truth. |
| Use Curbi from an agent client | MCP · Agent manifest | Read-only hosted MCP with explicit client compatibility. |
| Check a signed action record | Offline verifier · Proof draft | Open draft and test-key examples. Choose your own trusted keys. |
| Reproduce a broken tool workflow | Repair Lab · MCP Chaos | Included local fixtures and runnable regression artifacts. |
| Add checks inside a framework | Published adapters | Pinned framework tests with stand-in models and speakers; local source distribution. |
Use a concrete example.
Check a claimed refund against a supplied record, change the permission in a simulated action, run the published proof vectors, or replay the timeout repair fixture. Examples keep their test and simulation labels visible.
Inspect the contracts.
OpenAPI 3.1 describes the HTTP checker. MCP tool reference lists the hosted tools. Curbi Proof 0.1 describes signed records and local verification. The documentation catalog and llms.txt help agents retrieve published documentation.
Keep the roadmap separate.
Production signing is off; the published key set must be checked for current keys and validity dates. The reference proof log is in memory. Durable published logs, witnesses and production provider evidence are future work. None of the public tools provides a hosted execution gateway.
Run a check. Keep the result.
Check a reply Connect your agent →A pass means no configured rule fired. It grants no execution authority or provider-completion proof.