System

Settings

Workspace identity, the Grok Bot bridge, and developer data controls. Live data comes from the backend; demo data is opt-in and clearly labelled.

Operator

Integrations · Grok Bot

Sign in and connect the database to manage the Grok Bot bridge. Bridge tokens are never available in demo mode.

Interface

Data source

Live backend. Database status: checking. Currently reading 0 memories, 0 agents, and 0 projects.

Developer · Agent update protocol

Runtime

Local / Demo

Transport

Local / Demo

Queued events

0

Agents never touch screens directly. Every update is a validated envelope sent to the dashboard event service: the event is appended to the activity log, and any patch is upserted onto the referenced entity if its version is newer. Full contract lives in docs/AGENT-DASHBOARD-PROTOCOL.md.

  1. Identify yourself with a stable actor id and name.
  2. Send every update through the dashboard event service — never mutate UI state directly.
  3. eventId must be unique; duplicates are ignored, so replays are safe.
  4. Include eventId, occurredAt, actor, eventType, status, and correlationId on run events.
  5. Read the current entity version before patching; stale versions are rejected.
  6. Patch only the fields you own — never replace whole records.
  7. Emit started → progress (meaningful changes only) → completed or error for every run.
  8. Log approval.requested before waiting for a human, approval.resolved after.
  9. Never mark an app or runtime connected unless a real health check succeeded.
  10. Never delete or rewrite historical activity events — the log is append-only.

Event types

agent.registered · agent.heartbeat · agent.capability_verified · agent.started · agent.progress · agent.completed · agent.error · agent.status_changed · memory.created · memory.updated · memory.deleted · project.updated · project.created · skill.updated · routine.started · routine.completed · routine.error · application.connected · application.disconnected · task.queued · task.claimed · task.started · task.progress · task.completed · task.error · task.cancelled · approval.requested · approval.resolved · system.health

{
  "eventId": "3f9a1c2e-0d44-4f21-9d3e-1b7e0c9a5f10",
  "eventType": "agent.started",
  "occurredAt": "2026-09-10T09:41:00.000Z",
  "actor": {
    "type": "agent",
    "id": "research-agent",
    "name": "Research Agent"
  },
  "entity": {
    "type": "agent",
    "id": "research-agent",
    "name": "Research Agent"
  },
  "projectId": "prj-website",
  "status": "running",
  "title": "Research Agent started a run",
  "message": "Collecting sources on managed-agent positioning",
  "correlationId": "run-8c1f4b2a",
  "patch": {
    "status": "active",
    "lastActivity": "2026-09-10T09:41:00.000Z"
  },
  "version": 8,
  "metadata": {
    "trigger": "schedule"
  }
}

About

AgentNext Second Brain — the private operating system behind the public AgentNext site. ARMS model: Applications, Routines, Memory, Skills.