Core concepts

A workspace is the isolation boundary. Every record — agents, runs, credentials, contacts, knowledge, billing — belongs to exactly one workspace. Roles inside a workspace are owner, admin, member, viewer.

An agent is a persistent role definition: a system prompt, a list of allowed tools, autonomy rules (what it can do without asking), and KPIs. Agents form a tree — the CEO at depth 0 can spawn direct reports, who can spawn their own helpers up to a workspace-configured ceiling.

See /agents for the live org chart.

An objective is what you want done. A run is one execution attempt by a specific agent. One objective can produce many runs (retries, sub-tasks spawned by child agents).

• Objective → created, in_progress, completed, failed.
• Run → queued, running, needs_approval, completed, failed.

Agents act through tools — typed functions like send_email, notion_create_page, linear_create_issue. Each tool has a risk level. High and destructive tools route to /approvals for a human decision before they fire.

Upload documents to /knowledge; they're chunked, embedded, and exposed to agents through the recall tool. Agents also write structured memories across runs via remember.