The split
Foundry runs models. Kimss runs the operational layer around them: which workspace a call belongs to, whether that agent is still allowed to run, and a record of the requests that came through the gateway.
That is a control plane, not a second place to host weights. We do not resell inference. You bring the agents and the endpoint. OpenAI-compatible traffic can land at api.kimss.ai.
Room to build it
Microsoft for Startups Founders Hub gave us the Azure room to build this, with real infrastructure support rather than a slide. Thank you to Adir Ron, Amit Svarzenberg, and Moshik Shir. The sponsorship was how we got a secure platform standing. It is not Kimss selling compute on their behalf.
What production-grade means now
The product language has tightened since the first public note. Here is the version that matches what we ship:
- Workspace isolation for the tenant on the gateway.
- Identity on governed calls. Developer includes that path. Entra sign-in and SCIM provisioning are Enterprise, not part of the free tier.
- Metering in governed requests. Developer is 25,000 a month, then a hard stop. Production is $49 a month with 100,000 included.
- Azure Marketplace for the commercial path, beside direct signup.
Gateway-verified audit covers calls that passed the gate. It is not an immutable exhibit for usage somebody typed into a form.
Adapted from a LinkedIn note on 5 July 2026.