Agent engineering studio
We build agents that transact, work in the standards that let them interoperate, and publish the suites that prove an implementation is correct.
One discipline, three industries. Read the contract, build to it test first, prove it with a suite anyone can run.
Shopping and merchant agents over a real store.
Agentic checkout fails on protocol, not on prose.
Healthcare agents stand or fall on standards and interoperability.
Product
A phone number for a small business that answers every call.
It learns the business from its own website and documents, then handles inquiry, booking, returns and refunds, and hands to a person when it should.
In development. Proof of concept complete, early access opening to anchor customers.
We implement specifications until they break, report what broke, and send the fix. Public and countable.
65 merged pull requests across the specification, both official SDKs, the reference samples and the official conformance suite. 51 defect reports. Signature verification, the error envelope, idempotency, schema constraints dropped by code generation, webhook signing and delivery.
Open mandate constraints dropped by code generation before the checkout binding can enforce them. Versioned golden vectors with a drift canary against the reference.
Published without accepting contributions, so we ran its contract as tests against its own examples. A dozen departures found and reported. Twelve adopted by a maintained fork.
Check the claims before you talk to us.
Point it at any Universal Commerce Protocol server for a capability scoped report. No install.
The same suite on PyPI for your CI. 250 merchant checks, 44 agent checks.
Six real agent defects failing live. Conformance is not reliability.
137 numbered statements as pytest mixins, one report row each.
Claude Commerce Agents over Medusa v2. The adapter the reference does not ship.
The active protocols and implementations, kept current.
Every check is proven to catch the defect it was written for.
We watch it fail for the right reason, then we try to break the fix.
One report becomes a run of merged changes.
What we ran, what we skipped, and never certified.
A storefront about to expose an agentic interface, an agent that has to check out across platforms, or a healthcare integration that has to survive an audit.