Why an operating system
ROAM is not the agent.
It is the system the agents run inside.
Frameworks create agents. Harnesses run sessions. ROAM gives those agents the missing operating environment: shared state, durable memory, identity, rules, handoffs, trust signals, and operator command across the harnesses and tools you already use.
It has a substrate: the filesystem becomes the coordination bus every harness can use.
It has memory management: canon, journal, archive, decisions, and handoffs survive sessions.
It has identity: personas and authority boundaries persist when models or harnesses change.
It has a kernel: coordination, routing, escalation, and lifecycle happen in the system layer.
It has rules: approvals, boundaries, and audit records shape what agents can do.
It has a trust layer: EYDII watches behavior outside the agent runtime.
The architecture is the answer.
If this were only a dashboard, wrapper, or orchestrator, it would not need layers. ROAM earns the operating-system claim because the same system primitives support every agent: substrate, memory, identity, rules, coordination, trust, and harness independence.
Shown top-down from L07 to L00. The foundation starts at L00.
- L07
Harnesses & Roles
The agents and roles people see.
Claude Code, ChatGPT, Cursor, n8n, Ollama, local models, and custom harnesses can join the same organization without becoming the same tool.
- L06
EYDII — Trust Layer
Behavioral intelligence outside the work.
EYDII watches behavior, drift, scope movement, verification gaps, and agent health without reading prompts, code, messages, or artifacts.
- L05
Coordination Kernel
Routes work and enforces boundaries.
The system routes handoffs, resolves coordination events, reacts to trust signals, and keeps the organization moving without making a model mediate every step.
- L04
Rule Intelligence
What agents can do, when they must ask.
Approvals, escalation paths, authority boundaries, decision records, and audit events make autonomy manageable instead of invisible.
- L03
Identity & Privacy
Every agent has a durable identity.
Personas, authority, receipts, local credentials, and privacy boundaries stay separate from any single model session or harness.
- L02
Persistent Memory
Canon, journal, archive, decisions, handoffs.
The organization remembers settled decisions, active work, context, and supporting artifacts even when sessions end or models are replaced.
- L01
File System — The Substrate
The shared coordination bus.
ROAM uses the one primitive every harness can read and write: files. Tasks, messages, heartbeats, state, memory, and handoffs live in structured local files.
- L00
Hardware — Your Machine
Your hardware, your OS, your keys.
ROAM runs where the work already lives. The operating layer sits with your projects, credentials, memory, and agent work instead of becoming another cloud gate.
Primitives
The objects the operating system manages.
Boundaries
What ROAM is not.
Not a model.
Not a prompt wrapper.
Not a cloud gateway in the execution path.
Not a replacement for your harnesses.
Not an LLM coordinator for routine routing.
The practical test
Swap the harness. Keep the organization.
When a model, harness, or agent changes, the work should not reset. The memory, role, handoff state, approval history, and trust context should still be there. That is the difference between running agents in tabs and running agents inside an operating system.
No sign-up. Free up to three harnesses.