iiRecord
Agentic AI Atlas · observer-dashboard — real-time run observability
page:docs-user-guide-ecosystem-observer-dashboarda5c.ai
II.
Page reference

page:docs-user-guide-ecosystem-observer-dashboard

Reading · 2 min

observer-dashboard — real-time run observability reference

Docs(../index.md) › Ecosystem(./overview.md) › observer-dashboard

Pagewiki/docs/user-guide/ecosystem/observer-dashboard.mdOutgoing · 0Incoming · 1

DocsEcosystem › observer-dashboard

observer-dashboard — real-time run observability

**Package:** @a5c-ai/babysitter-observer-dashboard · **Version:** 6.0.0 · **Maturity:** GA

**The observer dashboard is a real-time observability UI for Babysitter orchestration runs. It streams run events over SSE and renders task progress, journal events, and orchestration state live in your browser.**

Because every run is event-sourced, the dashboard does not need to poll or guess — it simply consumes the same immutable journal the engine writes, as it is written.

---

On this page

---

What it is

Per its package description, it is the "real-time observability dashboard for babysitter orchestration runs." It is built on **Next.js** (the v6 announcement cites Next.js 16 / React 19), streams run events via **Server-Sent Events (SSE)**, virtualizes large run histories with @tanstack/react-virtual, and uses the Compendium design system. It renders task progress, journal events, and orchestration state in the browser.

---

How to launch it

The dashboard ships a bin and is also launchable from inside a harness:

bash
# In-session (Claude Code and other harnesses):
/babysitter:observe

# Via the genty runtime CLI:
genty observe

The underlying bin is babysitter-observer-dashboard (dist/cli.js). Source lives under packages/observer-dashboard/src/ (app/, components/, lib/, cli.ts).

---

How it works

Code
Engine appends events  →  immutable journal (~/.a5c/runs/<runId>/)  →  SSE stream  →  Dashboard

The engine writes journal events as the run progresses; the dashboard subscribes to the stream and renders them. Large histories are virtualized so the UI stays responsive on long-running or high-volume runs.

---

Who uses it

Anyone monitoring a live or historical Babysitter run — to watch convergence, see which task is executing, inspect the journal, or debug why a run stopped at a gate.

---

Next steps