TokenTelemetry
TokenTelemetry Docs

Quick Start

Launch TokenTelemetry for the first time, let it auto-detect your agents, and open the dashboard.

1. Start TokenTelemetry

The one-line installer launches TokenTelemetry for you on first install. To start it again afterwards, run the launcher from inside the tokentelemetry/ folder it cloned (there is no global tokentelemetry command):

cd tokentelemetry
./start.sh            # macOS / Linux
# start.bat           # Windows
# node bin/cli.js     # cross-platform

On first launch the launcher:

  1. Creates a Python virtual environment under backend/venv/ (takes ~30 seconds).
  2. Installs backend and frontend dependencies.
  3. Starts the FastAPI backend on port 8000.
  4. Starts the Next.js frontend on port 3000.
  5. Opens http://localhost:3000 in your default browser automatically.

Subsequent starts are fast — setup is skipped when nothing has changed.

2. Open the dashboard

If your browser doesn't open automatically, navigate to:

http://localhost:3000

You'll see the dashboard with KPI cards at the top (sessions, tokens, projects, cost) and charts below. If you haven't run any agents yet the counts will be zero — that's expected.

The dashboard is local-only by default

TokenTelemetry binds to 127.0.0.1, so http://localhost:3000 works only on the machine running it. Opening it from your phone or another computer is opt-in and token-gated — see Remote Access for the --host and SSH-tunnel options.

3. Run an agent session

Open a project in your terminal and start an AI agent session as you normally would:

# Example: Claude Code
claude

# Example: Gemini CLI
gemini

# Example: OpenAI Codex
codex

TokenTelemetry will detect the new session automatically. The dashboard refreshes every 15 seconds, so you'll see live token counts appear within a few seconds of starting the agent.

4. Explore

  • Click Traces in the sidebar to replay the full session step by step.
  • Click Projects to see per-directory activity.
  • Click Analytics to see token trends over time.

Auto-detection

TokenTelemetry looks for agent logs in their default locations:

  • ~/.claude/projects/ for Claude Code
  • ~/.codex/sessions/ for Codex
  • ~/.gemini/ for Gemini CLI
  • And so on — see Supported Agents for the full list.

No configuration is needed. If you keep agent logs in a non-default location, see Custom Data Directory — though that flag controls where TokenTelemetry stores its own state, not where it looks for agent logs.

Suppressing the auto-open browser

If you're running TokenTelemetry in a headless environment (a CI box, a VPS, inside an agent harness), set:

AGENT_HARNESS_NO_OPEN=1 ./start.sh    # or: AGENT_HARNESS_NO_OPEN=1 node bin/cli.js

The launcher will start both services without trying to open a browser.

Stop

Press Ctrl+C in the terminal where TokenTelemetry is running. Both the backend and frontend stop cleanly.

On this page