aitop

htop for AI agents. Find every AI coding assistant running on your machine. Spot zombies. Kill runaways. Reclaim memory.

$ aitop

  aitop — 14:32:08
  ──────────────────────────────────────────────────────────
  3 processes  ·  1.2GB RAM  ·  1 alert

  PID      TOOL           USER   MODE STATE UP      MEM     CPU  WORKDIR
  ──────────────────────────────────────────────────────────
  48291    Claude Code    dev    tty  sleep 2h15m   412MB   0%   ~/myproject
  48455    Cursor         dev    tty  sleep 1h02m   380MB   3%   ~/webapp
  31002    Aider          dev    bg   sleep 26h3m   445MB   0%   ~/old-project

  ✖ PID 31002 (Aider) running for 26h3m — likely zombie

Why

AI coding assistants run as background processes. They accumulate. A Claude Code session you forgot about yesterday is still eating 400MB. That Cursor window you closed didn't actually kill the process. Aider from last week is still hanging around.

htop shows them as generic node or claude processes. You have to know what to look for. aitop does that for you — it knows every major AI tool and shows you exactly what's running, for how long, and how much it's costing you.

Install

npm install -g aitop

Or run without installing:

npx aitop

Usage

# Snapshot — see everything right now
aitop

# Live mode — auto-refreshes like htop
aitop -w

# JSON output — pipe to jq, log to file, feed to monitoring
aitop --json

# Kill a specific zombie process
aitop --kill 31002

# Emergency: kill ALL AI processes
aitop --kill-all

# Skip Docker container scanning
aitop --no-docker

# Show child processes and node/python AI scripts
aitop --verbose

# Custom refresh interval (default: 5s)
aitop -w --interval=2

What it detects

aitop scans for all major AI coding tools:

| Tool | Provider | Process | |------|----------|---------| | Claude Code | Anthropic | claude | | Codex CLI | OpenAI | codex | | Cursor | Cursor | cursor | | GitHub Copilot | GitHub | copilot | | Aider | Aider | aider | | Cody | Sourcegraph | cody | | Amp | Sourcegraph | amp | | Windsurf | Codeium | windsurf | | Continue | Continue | continue | | Goose | Block | goose | | Tabby | TabbyML | tabby | | Ollama | Local | ollama | | LM Studio | Local | lmstudio | | Open Interpreter | Open Interpreter | interpreter | | GPT Engineer | GPT Engineer | gpt-engineer | | Devin | Cognition | devin |

With --verbose, it also scans node and python processes for AI framework patterns (LangChain, AutoGen, CrewAI, OpenAI SDK, Anthropic SDK).

Docker containers with AI-related names or images are detected automatically.

What it shows

For each AI process:

• PID — process ID (use with --kill)
• TOOL — which AI tool it is
• USER — who started it
• MODE — tty (interactive terminal) or bg (background/headless)
• STATE — sleep, run, zombie, stopped
• UP — uptime (color-coded: green <4h, yellow >4h, red >24h)
• MEM — memory usage in MB
• CPU — CPU percentage
• WORKDIR — which project directory

Alerts

aitop warns you about:

• Zombie processes — AI processes running >24 hours (red)
• Headless sessions — processes without a terminal >4 hours (yellow)
• Memory hogs — processes using >1GB RAM (yellow)
• Duplicate sessions — multiple instances of the same tool in the same directory (yellow)

JSON output

aitop --json

{
  "timestamp": "2026-03-20T14:32:08.000Z",
  "platform": "linux",
  "processes": [
    {
      "pid": 48291,
      "tool": "Claude Code",
      "provider": "Anthropic",
      "user": "dev",
      "cwd": "/home/dev/myproject",
      "mem_mb": 412,
      "uptime_sec": 8100,
      "cpu_pct": 0,
      "state": "S",
      "interactive": true
    }
  ],
  "containers": [],
  "alerts": [],
  "summary": {
    "total_processes": 1,
    "total_memory_mb": 412,
    "by_tool": { "Claude Code": 1 }
  }
}

Useful for piping into monitoring systems, cron alerting, or dashboards.

Use cases

Solo developer — "Why is my laptop fan spinning? Oh, 3 forgotten Claude sessions eating 1.2GB."

Team server — Run aitop --json in a cron job. Alert on Slack when AI processes exceed thresholds.

CI/CD — Check that no AI processes leaked from a previous pipeline run before starting a new one.

Agent orchestration — Monitor autonomous AI agents in production. Spot runaways before they burn your API budget.

Platform support

| Platform | Status | |----------|--------| | Linux | Full support (reads /proc for accurate data) | | macOS | Full support (uses ps + lsof) | | Windows/WSL | Works inside WSL, native Windows not yet supported |

Requirements

• Node.js 18+
• Docker (optional, for container detection)

License

MIT

Contributing

PRs welcome. If your favorite AI tool isn't detected, add it to AI_TOOLS in src/scan.js and open a PR.