🧹 RigSweep
Room for your agents. Free up RAM, CPU and disk on Windows dev machines cluttered by AI-agent sessions.
🌐 Homepage · ⬇️ Download · 📖 Antivirus FAQ
🇬🇧 English · 🇩🇪 Deutsch · 🇪🇸 Español · 🇫🇷 Français · 🇮🇹 Italiano · 🇧🇷 Português · 🇷🇺 Русский · 🇯🇵 日本語 · 🇰🇷 한국어 · 🇨🇳 简体中文 · 🇮🇱 עברית · 🇸🇦 العربية
📖 Why I built this
In the AI era, my laptop isn't running one thing anymore. On a normal day I have several Claude Code sessions in different terminals, a dev server per project, build watchers, and a browser full of docs, previews and dashboards. Every new session spawns more Node processes, and when I close a terminal some of them just keep running in the background.
It started to feel cluttered: fans spinning, RAM always near the top, and no clear picture of what was actually eating it. Task Manager just showed me a wall of node.exe and chrome.exe without telling me which session, which project, or whether anything still needed it.
So I measured my own machine (32 GB RAM) in the middle of a working day:
| 🔎 What I found | Count | RAM |
|---|
| 🌐 Chrome processes | 37 | 6.5 GB |
| ⬢ Node + Claude processes (agent CLIs, dev servers, the desktop app) | 14 | 2.5 GB |
| 🧩 Windows services running (telemetry, Xbox, search indexer…) | ~150 | — |
None of it was "a virus" or "junk". It was the normal leftovers of working with AI agents, plus Windows background stuff a dev box doesn't need. Existing "PC booster" apps either do scary irreversible things or know nothing about dev workflows. So I built the tool I wanted:
See what's actually using your machine, sweep what's stale, and keep every change reversible.
✨ Features
🧹 Dev Cleanup: stale dev processes in any stack
Finds dev servers, build daemons and AI-agent CLIs and groups each one with its whole process tree, project folder, age, RAM and listening ports.
| Stack | What it finds |
|---|
| ⬢ Node · Bun · Deno | dev servers (Vite, Astro, Next…), watchers, esbuild |
| 🐍 Python | uvicorn, Flask, Django, Jupyter, Streamlit, Celery, venv processes, Python MCP servers |
| ☕ Java | Gradle & Kotlin daemons, Maven, Spring Boot, Quarkus |
| 🟣 .NET | dotnet watch/run, Roslyn compiler server, leftover MSBuild nodes |
| 💎 Ruby · 🐘 PHP · 🦀 Rust · 🐹 Go | rails/ruby, php -S/artisan, cargo watch/run, go run |
| 🤖 Agent CLIs | Claude Code, Codex, Gemini CLI, opencode, Aider, Goose |
| 🔌 MCP servers | servers your agent started (@modelcontextprotocol/*, @playwright/mcp, mcp-server-*, FastMCP…) and left running after it exited, even behind a cmd /c npx launcher |
| 🌐 Automation browsers | Playwright / Puppeteer / --headless Chrome & Edge with their whole renderer tree. Your own browser is never listed |
| 🐧 WSL 2 / Docker | the Linux VM (vmmemWSL), shut down cleanly with wsl --shutdown, never pre-selected |
Python and Java also run ordinary desktop apps, so for those only processes with a clear dev signature are listed.
- 🧟 Orphan: its terminal was closed but it kept running. Pre-selected, even if it still holds a port (the classic zombie
localhost:3000).
- 🔌 MCP server: tagged on every row that is one, so you can tell an agent's leftover tool server from your own dev server.
- 👯 Older copy: an MCP server that's also running as a newer copy (e.g. under your current agent session) is marked, so the leftover one is easy to spot.
- 📈 Live meter: RAM and CPU are sampled every 2 seconds while RigSweep runs (window or tray). Right after a cleanup, manual or automatic, it shows how much RAM was actually recovered and CPU before → after, measured, not estimated.
- 🌡️ Temperature tracking: CPU and GPU temperatures live on the Dashboard with a 5-minute trend and peak, and a warning when the machine stays above 90 °C. GPU comes from the same counter as Task Manager; CPU from the ACPI thermal zone (no driver installed, no admin needed; approximate on many desktops). Click the tile to switch °C / °F.
- 💤 Idle: no CPU activity for a long time. Runtimes with no open port get pre-selected. An idle agent session in an open terminal is never pre-selected, because it's just waiting for you (and
claude --resume brings it back anyway).
How long counts as idle is set in Settings (in minutes, default 120; e.g. 5).
- 🧹 Auto-cleanup (off by default, in Settings): every minute RigSweep ends stale dev runtimes it has watched sit with no CPU for the whole idle period. Agent CLIs, the WSL/Docker VM and editor-owned processes are never auto-ended, and each ended tree is logged in History.
- 🟠🔴 Tray indicator: the tray icon (and the taskbar button while the window is open) gets an orange dot when stale dev processes hold ≥ 1 GB and a red dot at ≥ 2 GB. Hover the tray icon to see how much. Thresholds are set in Settings, where the dot can also be turned off.
- ♻️ Build daemon: Gradle, Kotlin and Roslyn servers are meant to linger. Ending them is harmless; they restart on the next build.
- 🔒 Locked: processes owned by VS Code, Cursor, JetBrains, Visual Studio… (your language servers) or by a desktop app are never killable from here.
- ✅ Nothing is ended without your confirmation, and PID reuse is checked before every kill.
以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗
