---
title: Dashboard tour
description: Every page of the ClawMetry dashboard, what question it answers, and where its data comes from — Home, Agents, Activity, Cost, Guard, Alerts and the rest.
keywords: ClawMetry dashboard, agent observability UI, agent monitoring dashboard tour
eyebrow: Dashboard
---

# Dashboard tour

The dashboard runs at `http://localhost:8900`. Its navigation is organised by
the question you arrived with, not by the subsystem that answers it.

## The map

**Top level — the questions you have most days**

| Page | Answers |
|---|---|
| [Home](/docs/dashboard/home/) | Is everything OK, at a glance? |
| [Agents](/docs/dashboard/agents/) | Every agent on this machine: what it runs, what it costs, is it alive, who owns it |
| [Activity](/docs/dashboard/activity/) | What are my agents doing right now, step by step? |
| [Cost](/docs/dashboard/cost/) | Where is the money going? |
| [Conversations](/docs/dashboard/conversations/) | What was actually said — across channels and runtimes |
| [Approvals](/docs/dashboard/approvals/) | What is waiting on me? |
| [Guard](/docs/dashboard/guard/) | What has gone off track, and can I stop it? |
| [Alerts](/docs/dashboard/alerts/) | Tell me when something breaks |

**Developer — [deep-dive views](/docs/dashboard/developer/)**

Flow, Models, LLM Context, Agent Graph, Tools, Context usage, Runtime extras,
Ask.

**Advanced — [everything else](/docs/dashboard/advanced/)**

Schedules, Memory, Notifications, Logs, Security, Tool permissions, Skills,
Self-Evolve, Version impact, and NemoClaw governance when a NemoClaw node is
present.

## Two things that apply everywhere

### The runtime filter

Most pages take a runtime filter. It is not cosmetic: session ids are namespaced
by runtime, so filtering to `codex` genuinely scopes the query rather than
hiding rows client-side. A page with no data for the selected runtime says so
explicitly rather than rendering an empty chart.

### Everything reads through the daemon

No dashboard page opens the DuckDB file. Each handler asks the daemon's
localhost query server, which is why the local dashboard and the cloud
dashboard render identical numbers from the same rollups — and why, if the
daemon is not running, pages are empty rather than stale.

```bash
clawmetry status      # the first thing to check when a page is blank
```

## Tabs that hide themselves

A tab that would be permanently empty for your setup does not appear:

- **NemoClaw** shows only when a NemoClaw node is present.
- **Runtime extras** shows only when the selected runtime exposes something the
  generic tabs cannot — Grok's upload manifest, Cline's rejected tool calls,
  Copilot's credit ledger.
- **Approvals** is meaningful only with an approval path configured.

This is deliberate. A blank tab is worse than a missing one: it looks like a
bug, and it teaches people to stop clicking.

## Getting the same data without the UI

Every page has an API behind it, and all of them are documented in the
[endpoint index](/docs/api/index/). The raw store is queryable through
[the query API](/docs/data/query-api/), and an agent can ask the same questions
over [MCP](/docs/mcp/overview/).