Project 01 · Systems architecture
Nomos. One layer for all AI work.
An operating layer between people and the AI systems they use. Work, memory and permissions live in one place, and the models underneath can be swapped out.
Visit nomos.myrakilabs.com (opens in a new tab)- Field
- AI operations
- Stage
- Build
- Started
- Jun 2026
- Review owner
- John Phelan, founder
The need
Where the work gets stuck.
Most people now use several AI tools, and each one keeps its own history and its own permissions. Context gets rebuilt by hand every morning, work restarts instead of carrying on, and nobody can say who approved what an AI did.
Adding another model makes that worse. Nomos puts one layer above the models that holds the work, the memory and the permissions, so you can change provider without losing any of it.
What we’re building
- 01
Projects and sessions
Work is organized into projects and sessions you can pick up again, with sources attached to each project and one history you can search across every model.
- 02
Multi-model routing
Cloud and local models are all reachable from one workspace. The type of task and how sensitive the data is decide which model runs. Tools connect through MCP.
- 03
Permissions and audit
Grants and an audit trail are in from the first release, not added later. Agents and tools act only under an explicit grant, and every action is recorded.
Review points
Where people approve the work.
- 01Agents run with a person in the loop by default
- 02Tool and MCP actions need an explicit grant
- 03Requests that need sign-off wait for a named approver
Progress
What’s changed so far.
Latest: Oct 2026
Version 0.1 alpha open by request to a small number of early users.
Jul 2026
First architecture note published: why we’re building an operating system for AI work.
Jun 2026
Project started.