P1
Modular
Each organ system is independently deployable. Adopt what you need — workspace only, brain only, or the full loop.
Open source · built in public
An AI-powered operating system for agentic teams.
Every person gets a structured agent workspace. The team gets one shared brain that learns how you work and helps you collaborate. MIT-licensed and self-hosted. You decide what leaves your machine.
$ npx aios-init my-workspace
$ cd my-workspace
$ aios push --tier team
2 Repos. One system.
3 Tiers. Nothing leaks.
100% Self-hosted. MIT-licensed.
THE PROBLEM
Working with agents is amazing. In a team, it breaks down fast.
Your organization already has a brain — scattered across Slack threads, stale docs, meeting notes, and the person who left six months ago. AIOS is the missing layer that makes collective intelligence queryable, actionable, and alive for agentic teams.
01
Context bloats.
Every agent's context balloons independently. No one on the team is working from the same picture.
02
Tools entangle.
MCPs, APIs, CLIs, and third-party sync tools pile up. Keeping them aligned costs more than the work itself.
03
Agents move 10× faster.
Contributors with agents outrun those without. The productivity gap creates chaos, not momentum.
04
The team loses the plot.
People tunnel into execution and drift from OKRs and customer needs. Speed in the wrong direction is waste.
Principles
Built for teams, not just individuals.
AIOS is a coordination layer — modular organs you can adopt piece by piece, with permission and privacy baked into the architecture from day one.
P2
Open source
MIT-licensed and self-hosted. Clone the workspace per person, run the brain on your Postgres. No cloud lock-in.
P3
Permissioned
Every action is policy-governed. Teams define what runs autonomously and what requires human approval.
P4
Privacy-first architecture
Individual agent data stays personal by default. The collective brain receives only what members choose to push.
The system
A workspace for individuals, a brain for the team.
Your workspace, your rules.
A numbered folder spine, governance rules, validators, and multi-agent harnesses — one per person. You decide what leaves your machine.
Shared memory, managed context.
Every workspace pushes tier-tagged context to one brain. Ask it questions in plain English across the team's shared memory. Self-host on your own Postgres.
Nothing leaks without a push.
Private → team → external. You promote content deliberately through the spine, and the brain rejects admin-tier at the boundary. Audience tags are first-class, not afterthoughts.
Automated workflows, delivered.
Turn meeting transcripts into decisions. Run weekly synthesis. Catch scope creep. Harnesses that ship deliverables, not chat logs.
All your tools in one place.
Slack, GitHub, Linear, Notion, Granola and more—wired into your workspace via MCP and CLI, with Team Brain sync where the integration supports it.
Yours to run, start to finish.
Open source under MIT. Self-host the brain, clone the workspace per person. No cloud, no lock-in, no per-seat pricing.
RUNTIME-AGNOSTIC
One interface for all your agentic work.
Go from juggling tools, windows, and workspaces to a single place to manage everything your agents do — and coordinate it across the team.
Works with the agents you already use
Claude Claude Code Cursor Codex OpenCode OpenCall Hermes
+ any MCP client
BYOA
Bring your own agent.
Plug any agent into your workspace. AIOS gives it structure, governance, and sync — without touching how the agent itself works.
MCP
Extend with what you already have.
Connect existing tools over the Model Context Protocol. The Team Brain is just another MCP server — your agents talk to it natively.
GUI
No terminal required.
Prefer a UI? Drive the same harnesses and shared brain from the dashboard — every capability surfaces without a single command.
TWO LEVELS, ONE LOOP
Every integration plugs into the same spine — whether it's your private workspace or the team's shared brain.
Workspace integrations
Each agent gets a numbered folder spine, governance rules, and access to your private context — only what you choose to expose.
Every item leaving your workspace passes one programmable governance gate. Tier rules and compliance policies — the EU AI Act, GDPR, and regional data residency — are evaluated on each push. Team and external content is approved and crosses to the Team Brain; private and admin content is rejected at the boundary with a 422.
Programmable governance
One gate on the sync boundary. Tier rules and compliance policies — EU AI Act, GDPR, data residency — run on every push. Team and external content crosses; private and admin stay local.
Build a skill or harness once in your workspace, ship it to the shared team repository, and every teammate adopts it automatically — no manual redistribution. The skill travels from you, to the team repo, out to the whole team.
Share a skill, the whole team adopts it.
Create a skill or harness once, ship it to the team repo, and everyone adopts it. The loop closes automatically — no manual redistribution.
HOW IT WORKS
One Team Brain
Every agent-user and data source feeds a single shared brain — knowledge, memory, reasoning, and policy in one place. A context engine finds the patterns; the brain surfaces the work and acts on it, getting sharper with every use.
Agent-users on Claude Code, Cursor and the AIOS Workstation send daily updates, and data streams from Slack, Notion, Linear, GitHub, Granola, Google and more are ingested, into one Team Brain holding knowledge, memory, reasoning and policy. A context engine beneath it finds patterns and closes the loop. The brain surfaces a kanban board, a KPI dashboard and a query interface, and takes policy-governed actions — spawning agents, executing tasks, and improving the team's shared harness of skills, hooks and guards.
AGENT-USERSAgent-User A
Agent-User B
Agent-User C
Slack
Notion
Linear
GitHub
Granola
Google
Team Brain
Context Engine
Kanban board
KPI dashboard
Query interface
Spawn agents
Execute tasks
Improve harness
The difference
Why not just use Claude and GitHub?
AIOS is the layer on top that helps to manage rapidly evolving context, harnessing and tooling. It wraps agents in a UI, provides shared memory, and unified governance, so your whole team can run them — not just the people who live in a terminal.
Claude Code + GitHub
AIOS
Who can use it
Engineers comfortable living in a terminal
Anyone — a UI puts the agent harnesses one click away
Agent harnesses
Wire up the prompts and glue yourself. Manage across multiple workspaces.
Create a shared harness for the entire team based on best practices.
Team memory
Per-repo, per-session, scattered across chats
One shared brain, queried in plain English
LLM cost & performance
Each person tracks spend and model choice alone, if at all
Per-member brain query spend in one dashboard today — unified cross-provider rollup (Anthropic, Cursor, etc.) on the roadmap
Sharing & privacy
All-or-nothing repo access
Tiered: private → team → external. Nothing leaks without a push
Structure
A blank repo and whatever conventions you remember
A numbered workspace spine, validators, and governance per person
Ownership
Your repos, your machines
Same — MIT-licensed, self-hosted, no new cloud
HOW INFORMATION FLOWS
You decide what leaves your machine.
Every file carries an access tier. Guards and validators enforce it on your machine, and the versioned brain-api contract is the only way across — private work is rejected at the boundary and never leaves. Team and external work syncs, and insight returns.
Each file in your workspace is tagged with an access tier. On your machine, guards and validators enforce those tags. The versioned brain-api is the gateway in the boundary between your machine and the shared side, backed by programmable governance — only team and external content crosses to the Team Brain; private content stays local. Shared signal and grounded answers return to you, always within your tier.
THE VISION Where we're headed
A closed loop for humans and agents.
Every context push makes the system smarter; every query pulls the right signal back to the person who needs it. The result is a coordinated AI system that orchestrates work for maximum team productivity — without sacrificing individual autonomy.
"What Claude Code did for individual agent productivity, AIOS does for team agentic productivity."
01 — The flywheel
The more context flows into the Team Brain, the smarter it gets. It learns your patterns, habits, and needs — so you build better skills, with more context, faster.
02 — Best patterns spread
Ship a skill or harness to the team repo and everyone adopts it. No re-inventing the wheel; no sharing prompt files in Slack.
03 — Codebase health
Measure agentic readiness, test coverage, and release cadence. Learn best practices on the fly from real instrumentation — not blog posts.
04 — Real-time coaching
Every AI session is evaluated for token efficiency and impact. Get targeted feedback on how to become a more effective agentic engineer.
05 — Toward orchestration
Agents take on execution — even running the board. Humans run orchestration. The division of labour shifts permanently.
06 — Learning journeys
Customised daily learning across 20+ job functions, in any tool. Coming next.
It starts with one user, grows to a team, and transforms a company. The flywheel of agentic transformation.
Honesty
The hard problems we're solving in the open.
A collective brain is not a solved problem. These are the design challenges we track publicly — and the reason architecture choices matter from the start.
Context overload
Too many signals, too much noise. Attention and prioritization are the hardest design challenge. A brain that pages you about everything is worse than no brain at all.
Contribution incentives
The system is only as good as what people push. Social design matters as much as technical design — tier tags and governance only work when the team participates.
Governance
Building a coordination system requires coordinating the builders. Early decisions set the cultural tone permanently — policy defaults and approval flows are not afterthoughts.
Trust in autonomous agents
Earned incrementally. Start narrow, demonstrate reliability on team-tier work, then expand scope. The action layer stays policy-gated for a reason.
The schema problem
What is the unit of knowledge? How information is represented decides whether the whole thing is queryable or just a fancier search index. OKF and tier-tagged items are our bet.
Staleness and decay
Knowledge expires. A brain that surfaces a stale decision as if it were current does real damage — so what to retain, what to archive, and when to forget is as hard as what to store in the first place.
ROADMAP
Where this is going.
AIOS is open source and shaped by the people who use it. If something on this list matters to you, open an issue or a PR.
SHIPPED
- Runtime-agnostic integrations (MCP + Plane/Linear/GitHub/Granola)
- Tiered access governance (private → team → external)
- Per-member token & cost tracking
- Agentic maturity instrumentation
- Dashboard task orchestration (brain → Linear projection)
- Open-source design system
IN PROGRESS
- Unified LLM cost & inference across providers (Anthropic, Cursor)
- Two-way PM sync (agents on the board)
- Slack bidirectional integration
PLANNED
- Finance & runway snapshots
- Agentic learning journeys (20+ roles)
- Public, live roadmap board
Getting started
Up in three commands.
# 1. Initialize your workspace
$ npx aios-init my-workspace
# 2. Run your first harness
$ aios run meeting-synthesis --input ./transcripts/
# 3. Push to the team brain
$ aios push --tier team
Agentic Transformation
An operating system for humans and agents working together.
One structured place where humans and agents collaborate — built through practice of transforming 100+ companies. It protects privacy, preserves context, and surfaces the best patterns from any contributor to the whole team. Your tools, skills, and memory in a single system that gets smarter the more you use it.
Get the workspace →