// answers

The words we use, explained.

Straight answers to the words of a collab OS — shared place, shared record, agents in the architecture.

// the category

What is a collaboration OS?

A collaboration OS is the operating layer where humans and AI work together as genuine workmates — same place, same shared record, same stakes. It is not a workspace with AI features bolted on, and not a chatbot with folders: the collaboration itself is the architecture.

The term exists because the current categories don't fit. Workspace tools organize your work but don't participate in it. AI assistants respond but don't keep a shared record. A collaboration OS is built for the relationship between the two — context earned once, held in the place, and worked by both sides.

Onnie is a collaboration OS: private Chats, shared Work, durable Artifacts, Agents on the team, and Knowledge you confirm — one place humans and agents share, not a filing system with a bot on the side.

What is context debt?

Context debt is the cumulative cost of re-explaining your work to AI — every new session, every tool switch, every re-opened project. You re-paste the background, re-describe the goal, re-orient the tool. Each instance costs minutes; across weeks it costs the momentum that separates projects that ship from projects that stall.

The debt compounds invisibly. Most people don't stop using AI because it's bad — they stop because the overhead of context swallowed the benefit. That exhaustion is context debt coming due.

Onnie eliminates it structurally: context lives in the place — shared Work, Artifacts, and Knowledge — not in a disposable chat window. You earn it once. It doesn't reset every morning.

What is the setup trap?

The setup trap is the belief — enforced by most productivity tools — that you must become a system architect before you can be a doer. Install the template. Configure the workspace. Build the automations. Connect the integrations. Only then, maybe, do the actual work.

The setup never ends, because the tools are built for maximum flexibility, which means maximum configuration. The people caught in it aren't doing it wrong — the mechanism is the problem.

Onnie is built against the trap: you start working in your first session, and the place forms structure around the work instead of before it.

// inside onnie

What is Work in Onnie?

Work is the room's shared multi-party record — digests, blockers, needs-you signals, and deliverables on one spine. A private chat is not Work; when something matters for the room, you track it as Work without freezing the thread. It is the shared spine humans and agents read together, not a classic project board sold as the product.

What are Artifacts in Onnie?

Artifacts are durable deliverables the place keeps — image, canvas, page, and table. They land on shared Work and in the library instead of vanishing as chat ephemera. Tables and pages are part of this family, not a separate product layer.

What are Agents in Onnie?

Agents are configurable AI teammates on your Work — with skills, tools, connectors, instructions, and memory. Run them from chat or on a schedule. They help move shared work and land durable Artifacts; they are workmates in the place, not a separate chat product bolted on the side.

What is Knowledge in Onnie?

Knowledge is the second brain in the place: atomic notes, typed links, hubs as lenses, and a Study gallery. Agents may suggest structure; humans confirm what enters the graph. It is not an Artifact — Artifacts are durable work outputs; Knowledge is confirmed memory you can reason over.

What are Routines in Onnie?

Routines are programmed recurring work that wakes itself — cadence, next fire, ghost runs — and moves shared Work without you having to re-ask. Public packaging is task-routines only: scheduled agent work that draws from the shared included capacity pool when it reasons.

What is included capacity in Onnie?

Each plan includes one shared workspace capacity pool for AI operators, agents, task-routines, connectors, and agent-authored tools — not a second tool meter and not a live chat taxi. Relative size (for example Pro is ~8× Free) and optional notional value describe the pool. When the pool is empty, AI reasoning hard-stops unless an admin enables extended capacity (opt-in real dollars) or adds seats.

Meet the collab OS these words come from.