Knowledge
FABRIC: plain markdown the team can edit, round-tripped into typed entities.
S
AI OPERATING SYSTEM · OPEN CORE
RealizeOS
A self-hosted team of AI agents that knows the business, remembers and works under human approval.
LIVEIN SHORT
RealizeOS is a self-hosted AI operations system: coordinated agents that share one knowledge base about the business, remember past work and act only within approval gates. Realization built it to run its own operations and publishes the core under a source-available licence.
- lines of Python, plus 11.5k of TSX
- 54.5k
- automated tests across 99 files
- 2,032
- MCP tools exposed to other agents
- 24
- LLM providers, routed by task and cost
- 4
CASE STUDY
Problem. Approach. Outcome.
- The problem
- Most teams adopted chatbots, not an AI operation. Every chat starts from zero, work isn’t followed through, and the knowledge lives with the model vendor.
- The approach
- Keep the durable asset — knowledge, history and identity — in files the business owns. Everything else is a swappable adapter: models, agent runtimes, channels and interfaces.
- The outcome
- Version 5.6 runs Realization’s research, documentation and coordination daily. The core is free to self-host; setup and adaptation are available through Advisory.
HOW IT WORKS
Context → coordinated action.
Six hard problems, solved as one stack.
Memory
Synapse: a four-tier index that recalls the cheapest context first.
Missions
An eight-state engine that plans, routes and tracks the cost of every step.
Governance
Six roles, injection guards, audit logs and human approval on consequential actions.
IMAGES
The work, shown.
Note. Source-available under BSL 1.1. Counts are measured from the repository.
Source. RealizeOS project overview v5.6.0, September 2026.
RELATED FIELD NOTES