Listen
0:00 / ~2:15
Loading workspacePreparing your latest view
The in-app conversational agent and its tooling surface. Endpoints cover sessions, custom tools, async jobs, mentions, attachments, feedback, and an undo log for any side-effects the agent took.
/api/ai-agent/conversations/api/ai-agent/conversations/api/ai-agent/conversations/{id}/api/ai-agent/conversations/{id}/api/ai-agent/conversations/{id}/export/api/ai-agentPOST /api/ai-agent is the streaming chat endpoint. Pass { conversationId, messages, tools? }and stream the SSE response.
/api/ai-agent/custom-tools/api/ai-agent/custom-tools/api/ai-agent/jobs/api/ai-agent/jobs/{id}Long-running operations the agent kicks off (bulk task updates, spec ingestion, large reports) are tracked as jobs.
/api/ai-agent/audit/api/ai-agent/undoEvery write the agent performs is journaled. Undo replays the inverse operation up to a configurable retention window.
/api/ai-agent/feedback/api/ai-agent/feedback/analytics/api/ai-agent/preferences/api/ai-agent/preferences/api/ai-agent/personas/api/ai-agent/personas/api/ai-agent/personas/{id}/api/ai-agent/personas/{id}/api/ai-agent/personas/{id}/api/ai-agent/personas/{id}/members/api/ai-agent/personas/{id}/members/api/ai-agent/personas/avatar/generatePersonas carry an optional allowed_scopes array (NULL = inherit everything), an optional model override, and an access-controlled member list. Create / update / delete / members-PUT all require workspace-admin; GET filters the catalog to personas the caller can actually reach. Avatar generation hits an image model and returns a JPEG data URL — the prompt is persisted so the avatar can re-render on edit.
/api/agent-alerts/api/agent-alerts/briefing/api/agent-alerts/scan/api/agent-alerts/dismissAlerts are produced by a background scan over tasks, deploys, and discussions. /briefing returns the prioritized morning summary.
/api/ai-agent/schedules/api/ai-agent/schedules/api/ai-agent/schedules/{id}/api/ai-agent/schedules/{id}/api/ai-agent/schedules/runsA schedule is a recurring prompt the agent executes on a cron. Times are UTC. autonomy_level is none (a single text reply, the default) or safe (the full tool loop with external sends, deletes, and payments stripped). /runs returns run history — admins see the whole workspace, members see schedules they created.
/api/ai-agent/action-policies/api/ai-agent/action-policies/api/ai-agent/action-policies/api/ai-agent/escalation-policy/api/ai-agent/escalation-policy/api/ai-agent/guardrails/api/ai-agent/guardrails/api/ai-agent/guardrails/api/ai-agent/guardrails/api/ai-agent/activity-reportAll admin-only. Action policies govern six risky categories (external_send, finance, destructive, membership, deploy, external_write) with a mode of allow / allow_budget / approve / block. A category with no policy row parks for approval — absence is fail-closed. A persona-scoped row shadows the workspace row. Guardrails and the escalation policy are keyed by ?id= query param on write. The activity report (?days=N, max 90) rolls up executed / parked / blocked / failed / undone actions.
/api/ai-agent/share/api/ai-agent/share/{id}Shares one message or a whole conversation. Destination knowledge saves it as a knowledge document; snapshotcreates a snapshot that can be DM'd to workspace members and, with visibility: "public", published to a public link.