Skip to content

The 8 organ systems

The AIOS Team Brain is organized as eight organ systems — independently deployable capabilities that together form a collective intelligence layer. This page adapts the architecture model from Chetan’s product deck for the public OSS audience.


SymbolMeaning
✅ ShippedMVP in the current tagged release
🟡 PartialSchema or engine exists; gaps remain (see notes)
⬜ PlannedOn the roadmap; not yet in production code

Everything the team knows, structured and retrievable. Items land in Postgres with full-text search; the query layer assembles context for natural-language answers.

Where: items table, FTS indexes, lib/query in the Team Brain.


Pipes from workspace pushes and external sources into the brain. The CLI sync path and Python sidecar handle tier-tagged writes at the API boundary.

Where: lib/ingest + ingestion/ sidecar. Eight source types in MVP.

See also: Brain API v1 for the sync contract.


Decides what surfaces and what stays quiet — signal over noise so the brain informs without overwhelming. Retrieval and ranking logic exists; prioritization heuristics are still evolving.

Where: lib/query/retrieve.ts in the Team Brain.

Related landing topic: Honesty — context overload (hard problem #1).


Policy-gated agents that act on behalf of the collective. The schema and API seam exist; a full autonomous runner is not wired yet.

Where: lib/actions, actions table, POST /api/v1/actions.

See: Harnesses for how workspaces orchestrate agent work today.


Who belongs, with what role and permissions. Teams, members, and per-member API keys are the root of access control.

Where: teams, members, api_keys tables.

See: Access tiers in the glossary and the tier model in Brain API v1.


The constitutional layer — what’s autonomous, what needs approval, who decides. Engine and schema exist; dashboard UI and write-path enforcement are not complete.

Where: lib/policy, policies, approval_requests tables.

Principle: Permissioned (landing page).


Immutable record of every action, query, and ingestion event. Append-only, trigger-backed — the trust layer for a shared brain.

Where: audit_log table in the Team Brain.


Closes the loop so the brain learns from outcomes over time. PM sync, work events, and codebase analytics feed progression signals; full closed-loop learning is still maturing.

Where: work_events, lib/pm-sync, codebase scan ingest.

See: Agentic Engineering maturity for individual and team instrumentation.


Workspace pushes (tier-tagged)
Ingestion ──► Knowledge repository ◄── Context management
│ │ │
│ ▼ │
│ Query / Surfaces │
│ │ │
▼ ▼ ▼
Audit log ◄── Policy engine ──► Action layer
│ │
└──────── Feedback loop ◄──────┘

For deployment and operation, start with Team Brain and Your Workspace.