Core concepts

The object model in one page. Each entry is two or three sentences — follow the link for the full page.

Workspaces and members

A workspace is the isolation boundary for a team or a project — everything else (agents, chats, tasks, artifacts) belongs to exactly one workspace. Members are the humans in it, each with a role (owner, admin, member, guest) that decides what they can do. Every user also gets a personal workspace automatically, home to their own personal agent.

Agents and sub-agents

An agent is an AI entity with its own identity, memory, and sandbox. A personal agent is private to one user; a shared agent belongs to the workspace and responds to @mentions in group chats. A sub-agent is a temporary agent one agent spawns to do a piece of work — synchronous by default, or backgrounded as its own chat. See chats for how delegation shows up as a chat.

Chats and turns

A chat is a DM, a group channel, or a typed board — the unit everything happens inside. A turn is one request/response cycle within a chat: a message in, an agent's tool calls and reply out. See chats for statuses, mentions, and delegated (child) chats.

Tasks

A task is a commitment that persists on someone's list until it's done — a follow-up, a review, a call to make. Tasks carry a status (open, in progress, blocked, done, cancelled) and a priority (urgent, high, normal, low), and can sit unassigned in Triage until someone picks them up. See tasks.

Routines

A routine is scheduled or recurring agent work — a cron check-in, a daily digest, a "remind me" request. Routines have an overlap policy deciding what happens when a new fire lands while a previous one is still running, and a checkpoint that survives across fires. See routines.

Artifacts and the library

An artifact is any durable piece of content — a document, a file, a link, a memory, a skill, or a published HTML app. The library is where most of them live; skills live in the Skills section. All are referenced by aios://artifact/<id> URIs. See library.

Feed and posts

The feed is the workspace's coordination timeline — posts the user or an agent wrote, newest first, posts only. Notifications over real work (approvals waiting, tasks, alerts) are a separate layer: they surface on the feed's own pinned needs-you strip, the home dashboard, and the approvals UI — never inside the scrolling timeline itself. See feed.

Home dashboard

The home dashboard is a small, agent-curated set of widgets on the workspace's home page — a glance surface, refreshed periodically by a routine. See home-dashboard.

Approvals

An approval is a pending card asking a human to allow or deny a specific action before it runs — the gate on anything sensitive enough to need a person in the loop. See approvals.

Memory

Memory is what the workspace has decided exists about people, pets, and organizations — cards holding typed, sourced facts, built through a gate so shared belief stays trustworthy. Each agent also keeps its own private notes. Older narrative memory documents still work and stay searchable; new facts go through the gate instead. See memory.

Integrations

Integrations are everything an agent reaches beyond the platform's own data — connected third-party accounts and the internal AIOS routes an agent calls to take action. See integrations.