Agenttop

agenttop

  htop for AI coding agents.
  See exactly where your Claude, Cursor, Kiro, Copilot, and Codex tokens are going — and what it's costing you.

  Install ·
  What you get ·
  vs other tools ·
  How it works ·
  AI Analysis ·
  Roadmap

Why this exists

I was using Claude Code, Cursor, and Kiro every day. Each of them stores usage data locally — JSONL logs, SQLite databases, workspace state — but none of them show you the full picture. I had no idea:

- Which project was burning most of my tokens
- Which model I was over-using
- Whether my prompts were causing correction spirals (they were)
- How much of my Cursor tab output was actually accepted

So I built agenttop. One view across every tool. Real numbers computed from actual tool-call data, not keyword guessing. Optional AI analysis that tells you *specifically what to change*. All local. Nothing leaves your machine.

Install

One line. Installs everything, asks what to launch (web / TUI, real data / demo).

Skip the menu and jump straight into a mode:

After install:

Requirements: Python 3.10+, git. No Docker. No API keys needed. macOS, Linux, Windows (WSL).

Keyboard (TUI): d dashboard · s sessions · e explorer · a analysis · k graph · 1-4 time range · q quit

What you get

Terminal dashboard

7 panels, updates live:

- Cost by project — which repo is burning your money
- Cost by model — opus vs sonnet vs haiku split
- Daily cost — 30-day histogram with total / avg / peak
- Hourly activity — when you actually work
- Activity breakdown — coding / debugging / testing / exploration %
- Tools — per-tool sessions, tokens, cost
- One-shot rate — % of edits that pass without retry

Web dashboard

Three tabs:

- Overview — force-directed knowledge graph (D3), model usage (input/output/cache), hourly activity,…