CellOS
The missing infrastructure layer for human-AI coordination. Define how humans and AI agents work together with formal roles, scope boundaries, and accountability chains.
One cell. Five roles.
Five humans, flat.
Every CellOS deployment reduces to this structure: five steward roles, each owned by a human, in a flat hierarchy. No boss node. Each steward runs AI agents beneath them, scoped to sub-functions - and an agent team can mirror the same five roles, one level down. The structure is fractal.
Hover a steward to read its mandate. Resomnium itself runs as cells of this shape.
The problem.
Organizations are deploying AI agents without coordination infrastructure. There are no standards for how agents interact with humans, no formal boundaries for what agents can and cannot do, and no audit trails for accountability.
Existing agent frameworks (LangChain, CrewAI, AutoGen) handle execution but not governance. They tell agents what to do, but not what they're allowed to do, who's accountable, or when to escalate to a human.
No scope enforcement
AI agents act without boundaries. No formal definition of what they can and cannot do.
No accountability chains
When an agent makes a bad decision, there's no trail showing who was responsible or how to escalate.
No coordination protocol
Handoffs between humans and AI are ad hoc. No formal structure for when AI should defer to human judgment.
Three layers of coordination.
CellOS provides the governance layer that sits on top of any agent framework, defining who can do what, who's accountable, and what happens when things go wrong.
Cell Schema
JSON Schema specification for defining organizational cells. Roles, participants, scope boundaries, escalation rules, and handoff protocols — all in a machine-readable format.
Coordination Engine
Python runtime that enforces cell definitions. Scope checking before every action, automatic task routing, escalation when agents hit boundaries, and a complete audit trail.
Diagnostic Toolkit
Assessment engine that analyzes cell definitions for completeness, safety gaps, and coordination risks. Scores your organization's AI readiness and provides actionable recommendations.
Three steps to governance.
Define your cell
Write a YAML file describing your team: who's in it, what roles they play, what each role is allowed to do, and what happens when something goes wrong.
Enforce at runtime
Before any agent takes action, CellOS checks it against the scope boundaries. Forbidden actions are blocked. Actions requiring approval get routed to the right person.
Audit everything
Every action, escalation, and handoff is logged. Know exactly who did what, when, and whether it was within bounds.
from cellos import (
load_cell_from_yaml,
CellRuntime,
Coordinator,
)
# Load your cell definition
cell = load_cell_from_yaml("team.yaml")
# Create a runtime
runtime = CellRuntime(cell)
# Check scope before any action
check = runtime.check_scope(
"ai-agent-1",
"send_email to client"
)
if not check.allowed:
print(f"Blocked: {check.reason}")
# -> "Action 'send_email' is
# forbidden for role 'Analyst'"
# Create and route tasks
task = runtime.create_task(
"Analyze Q1 revenue data"
)
routed = runtime.route_task(task.id)
# -> Automatically assigned to the
# best-fit participant
# Full audit trail
print(runtime.audit.summary(cell.id))Works with your stack.
CellOS is the governance layer, not a replacement for your agent framework. It sits on top of whatever you already use.
Google A2A
Agent-to-agent protocol
LangChain
Agent orchestration
CrewAI
Multi-agent systems
AutoGen
Conversational agents
Built from real operations.
CellOS isn't theory. It's derived from the Cell Framework that Resomnium uses to run its own operations — a 5-person cell structure with five steward roles (Clarity, Execution, Narrative, Access, Governance) where humans and AI agents coordinate daily.
The framework has been tested in production. CellOS makes it open-source and machine-readable so any organization can adopt it.
Learn about the Cell Framework →
The Cell Framework.
The Cell Framework is the organizational design model behind CellOS. It defines how small, autonomous teams of humans and AI agents coordinate through formal roles, clear scope boundaries, and structured accountability. CellOS is the software implementation that makes the Cell Framework machine-readable and enforceable.
Five steward roles
Every cell operates with five distinct steward roles, each with defined responsibilities and scope boundaries:
Clarity—Sets direction, owns final decisions
Execution—Coordinates tasks, manages delivery
Narrative—Handles communication and distribution
Access—Manages relationships and partnerships
Integrity—Ensures quality, compliance, and accountability
Flexible cell structures
The framework scales from solo operators to full teams, with each configuration balancing human judgment and AI capability differently:
Single Player—One human, multiple AI agents
Cell of 3—Small team with shared stewardship
Cell of 5—Full cell with dedicated steward roles
Startup Cell—Adapted for early-stage teams
Legacy Upgrade—Transitioning existing organizations
Deep dives on Substack
The Cell Framework is documented in detail across three article series exploring the future of organizational design, roles, and skills in the AI age.
Company of the Future
Jobs of the Future
Skills of the Future
Where we are.
Want to use CellOS in your organization?
We're looking for early adopters and research partners. Whether you want to contribute to the framework or pilot it in your team, let's talk.