← Back to Discover

token-meter

A terminal tool that watches an AI coding agent's local session files and shows live token use, estimated cost, messages and tool calls, as a full dashboard or a one-line status bar.

Observability & evalsCommand-line tools
Visit website ↗

What it does

token-meter is a command-line monitor for AI coding agents. While an agent runs, it shows token consumption for each turn, a running cost estimate, timestamps on each message, and which tools are being called and when. You start it with npx or install it globally with npm. It has two modes. The full dashboard takes over the terminal, and the recommended status bar mode shows a single line at the bottom with total tokens, estimated cost, message count, tool calls, sessions detected and the last event. Options let you pick an agent, a refresh interval or a specific session. Supported agents are Kimi Code, Claude Code, Codex and OpenCode. According to the README, the tool reads local session files as they are written, makes no network requests, never modifies those files and needs no API keys. It is MIT licensed, and the README points to a related tool, agent-trace, for analysis after a session.

Capabilities

  • Live token and cost readout during an agent session
  • Full dashboard mode or a one-line status bar
  • Timestamps on each message and tool-call tracking
  • Supports Kimi Code, Claude Code, Codex and OpenCode
  • Runs locally with no API keys or network requests

When to use it

  • You want to see what an agent session is costing while it runs
  • You run Claude Code, Codex, OpenCode or Kimi Code and prefer a terminal view
  • You want a monitor that stays local and needs no keys