Documentation

Overview

Navigate the visual layer on top of TraceLLM.

Overview

The TraceLLM dashboard is a web-based observability interface that surfaces every trace recorded by your LLM stack. It runs at localhost:3000 by default and requires the backend (started via tracellm start) to be running.

Dashboard Layout

The dashboard is organized into five primary views accessible from the sidebar:

ViewDescription
OverviewSummary of total traces, active projects, recent activity, and system health
TracesBrowse, filter, and inspect individual trace records with full step detail
AnalyticsAggregated metrics: latency, token usage, error rates, and trends over time
Live LogsReal-time event stream of trace arrivals via WebSocket
FailuresDedicated view for failed traces, retries, and slow requests

Info

The dashboard connects to the backend API at http://127.0.0.1:8000 and subscribes to WebSocket events at ws://127.0.0.1:8000/ws. Both addresses are configurable via environment variables.

How It Works

  1. Traces arrive from instrumented code (decorators, CLI, or integrations) and are persisted to MongoDB.
  2. A WebSocket event (trace.created) is broadcast to all connected clients.
  3. The dashboard receives the event and updates its trace list in real time without polling.
  4. Clicking a trace opens the inspector, which fetches the full trace document from GET /traces/{trace_id}.
  5. The analytics view queries GET /analytics to render aggregated charts and breakdowns.

First Steps

To start using the dashboard:

1

Start the stack

Run tracellm start in your terminal. This launches the FastAPI backend and the WebSocket server.
2

Generate traces

Run tracellm trace "your prompt" or use the @trace decorator in your application code.
3

Open the dashboard

Navigate to http://localhost:3000. Traces appear automatically as they are created.

Common Errors

ErrorCauseFix
Dashboard shows offlineBackend not running or port mismatchRun tracellm start and verify the API is on port 8000
No traces appearNo traces recorded or MongoDB is emptyRun tracellm trace to generate a trace
WebSocket disconnectedBackend restarted or network issueThe dashboard auto-reconnects — wait a few seconds
Blank page on loadFrontend build issue or missing dependenciesCheck the browser console for errors and restart the stack