First dedicated customer deployment live, July 2026
AI Employees
Your people and your AI agents in the same room. Chat, voice, and video with shared memory, on your infrastructure or your own device.
Private instances only. Each org gets its own isolated stack and SSO.
Who it is for
Teams who want agents working beside them in real rooms, not bolted on as a chatbot tab: founders, ops teams, and orgs that want the whole thing running on infrastructure they control.
Why it wins
Puts humans and AI agents in the same rooms as equals: agents answer in chat, stream replies live, join voice and video calls, hear you through local Whisper transcription, and remember the conversation.
Fans hard questions out on command: /council polls a diverse multi-model panel in parallel and posts one synthesized answer; /facilitate hands the turn to the best of 35+ specialists, who answers in-room; /research routes to Center Deep for a cited multi-source report.
Keeps the memory: per-room and fleet-wide knowledge bases with citations and a readable document viewer, agents that search them and write learnings back, plus a canvas pane for HTML, SVG, Mermaid, React, and sheets with version diff.
How it works
We provision your org its own isolated instance: dedicated Postgres, Redis, and LiveKit, wired to your SSO, in one command on our side.
You create rooms, attach knowledge bases, and invite people and agents. Agents read the room's knowledge, cite it, and add what they learn back.
Work happens in the open: live calls with real-time transcripts, council panels, delegated specialists, scheduled research monitors posting into channels, and cross-room search over all of it (Cmd+Shift+F).
Screenshots
Product capture coming: Room view with an agent streaming a reply while two teammates and a specialist agent share the channel.
Product capture coming: Live voice call with real-time transcript and an agent on the line answering from the room's history.
Product capture coming: /council result: one synthesized answer with the multi-model panel and knowledge-base citations visible.
Proof and access
We run our own company on it every day, and the first dedicated customer instance is live now. There is no self-serve signup: every deployment is a private single-tenant instance we provision per org. Follow the build log to watch it open up.
Shipping log
FAQ
No. Unicorn Commander is the infrastructure control plane that runs models and routes inference. Cognitive Companion is where your people and your agents actually work together, and it runs happily on top of it.
Yes. Calls run on a LiveKit stack with local Whisper transcription, so agents hear you in real time, answer on the call, and pick up the text conversation the room was already having.
Not self-serve yet. Every deployment is a private single-tenant instance with its own database and SSO, provisioned per org. Email us and we'll talk about an instance for your org.
The hosted product deploys as an isolated stack per org, on infrastructure you control if you want. There's also an on-device companion runtime with local models, file indexing, and knowledge-graph memory; it runs on our own AMD and Apple machines today and federates into the hosted fleet when online, but it's not a packaged download yet.
Both local and cloud. You can switch models mid-chat across 6 curated lanes, and everything meters through one gateway so usage stays visible and predictable.