AI OPERATING SYSTEM · OUR ENGINE

RealizeOS

A self-hosted team of AI agents that knows the business, remembers and works under human approval.

Operators connecting field equipment beside a tablet and process map

IN 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.

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. It is the engine inside our own operations and our client AI work, not a product we sell on its own.

HOW IT WORKS

Context → coordinated action.

Knowledge, memory, tasks and approvals in one system.

Knowledge

FABRIC: plain markdown the team can edit, round-tripped into typed entities.

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.

RealizeOS Missions dashboard showing mission totals, tracked cost and a completed property-search mission — Missions: goals planned into steps, with cost tracked per step · demo data

RealizeOS Dream Inbox listing knowledge updates proposed by background agents, each waiting for approval — Dream Inbox: agents propose knowledge updates, a person approves · demo data

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

The thinking behind it.

AI in practice

AI in the office: start with knowledge

Most teams use AI for small tasks. The value starts when the office's knowledge is in order.

AI in practice

Own the heart: local-first AI operations

Models, runtimes and tools will keep changing. Your business knowledge should not change hands with them.

NEXT STEP

Want an AI operation that belongs to you?

Or book a 20-minute intro call with Asaf.

Talk about a setup