New reference build shipped July 2026
Platform
A meeting recorder that physically cannot phone home. Private meeting intelligence on a box you own, sitting in your building.
Who it is for
Individuals, small and medium businesses, clinics, law offices, agencies, and boardrooms. Anyone whose meetings must never leave the room, and who wants an appliance rather than a project. Larger teams scale it by federating API and GPU nodes behind the same box.
Why it wins
Records, transcribes, and summarizes meetings entirely on the box. Audio in, finished minutes out, and no cloud path for anything to leak through. This is the proven, shipping configuration.
Can also be configured to run Unicorn Commander, the full platform. An external GPU over OCuLink is what makes that mode genuinely comfortable. Said plainly: in that configuration most inference goes out to an LLM API you choose rather than running on-device. The meeting appliance mode is the one that never phones home.
Joins your network as an always-on node: OpenAI-compatible endpoints, studio-grade text-to-speech, and room for background agents that keep working after everyone goes home.
How it works
Plug in the box. The reference build is a GMKtec NucBox K11: Ryzen 9 8945HS with an XDNA NPU and a Radeon 780M iGPU, 96GB of unified memory, tuned and loaded before it reaches you.
Record or import meetings. Transcription, speaker notes, and summaries run on the appliance itself, powered by the same Meeting-Ops software we sell on its own.
Grow when you're ready. One box serves an individual or a small team on its own. Add an external GPU over OCuLink for heavier local work, or federate it into Unicorn Commander alongside API and GPU nodes and it will carry an enterprise. Either way, the mode you're in is stated plainly, on the box and by us.
Screenshots
Product capture coming: Meeting summary produced entirely on the appliance.
Product capture coming: The GMKtec NucBox K11 reference unit with OCuLink eGPU dock.
Product capture coming: Failover dashboard for the suite voice endpoint.
Proof and access
Built to order. The reference unit runs in production inside our own fleet today. We assemble and tune each one deliberately rather than holding stock, so it starts with a conversation about what you need it to do, not a cart button. No published price yet, and we won't invent one.
Two honest modes. As a meeting appliance, nothing leaves the box. Configured to run the full Unicorn Commander platform, most inference goes out to an LLM API you choose. The settings page tells you which mode you're in, and so do we.
Shipping log
FAQ
Yes, built to order. We don't hold stock: each unit is assembled and tuned deliberately, so it starts with a conversation about your workload rather than a checkout page. There's no published price yet and we won't invent one. Tell us what you need it to do and we'll tell you what it takes.
In the meeting appliance configuration, yes, structurally: recording, transcription, and summaries all run on the box, and you can watch the network tab prove nothing leaves. If you configure it to run the full Unicorn Commander platform instead, most inference goes out to the LLM API you chose. That mode is opt-in and clearly labeled.
No. Local models run on the AMD Radeon 780M iGPU built into the mini-PC. An OCuLink dock lets you add an eGPU if you want one; our reference build uses one there for text-to-speech.
The Ryzen 9 8945HS includes an XDNA NPU, and we've built a bespoke NPU speech-to-text stack for it. Restoring that onto the current build is phase 2 on the roadmap; today's shipped workloads run on the iGPU and eGPU.
UC-1 federates into Unicorn Commander as a first-class inference node. Your apps call one logical model name and the platform handles routing, metering, and failover, whether the answer comes from UC-1 or another node.