Semurg BOS
One server
Everything the demo runs on sits on one rented server in Chicago, and this site's two forms end up there too. This is what is on it.
What runs on the server
Chicago · Cherry Servers · AMD EPYC 9554P · no GPU
BOS
The application and its windows. This is the demo at demo.semurg.org.
Semurg
The memory: your files, the facts in them, the links between them, and search. It is this server's own, not a service somewhere else.
The AI model
Gemma 4 (26B), an open model, run on the server's own processor. The model sends nothing visitors ask to an AI company.
The two forms
"Request demo access" and "Join the waitlist" on this site go straight into BOS.
What is not on it
- No separate database
- No outside AI for visitors' questions
- No form service
Under the hood
For the engineers.
One engine
Semurg keeps your records, the links between them, search by meaning and by time in one engine. Agents read and write it through one standard tool interface, MCP. No separate database to install or tune.
The same memory, for your agents
Semurg is the memory under BOS, and AI agents can work from it too. Our own team builds BOS this way.
Any agent tool
Agents reach the memory through MCP, the open standard for AI tools, from any agent tool that speaks it.
A memory that lasts
It searches by meaning and follows the links in the graph, so it brings back what bears on the task at hand, however long ago it was stored. That is how our agents keep working on a large code base across sessions.
Agents that talk
Agents send each other tasks and answers through Semurg's own messaging, built on telecom-grade technology.
Expert spaces
An Expert space holds what one role has learned, so a fresh agent can load it and get to work at once, aware of the whole team around it.