Void

Reclaim your disk space

Void cleans up after your AI agents — and your agents can call Void themselves. Worktrees, transcripts, local models and the build artifacts behind them, across 15 ecosystems, with exactly what each cleanup will do shown before it runs.

$ curl -fsSL https://raw.githubusercontent.com/eladbash/void/main/install.sh | sh

Native on macOS, Windows and Linux · MIT licensed · runs locally · no account, no telemetry

Void's results screen: build artifacts grouped by ecosystem, each row showing its full path, age, size and risk level.
Results grouped by ecosystem. Every row shows the path, how stale it is, and what cleaning it would actually do.

Built for people who are careful with their machine

Every project you touch leaves something behind. Void answers where it is, how much it costs you, and whether it is safe to remove.

One scan, everything at once

A multi-threaded walker sweeps your scan roots in parallel and streams results as it finds them. No waiting for a total before you can start reading.

The exact command, up front

Void never hides what it is about to run. Whether it is cargo clean, docker builder prune or a plain delete, you see the literal command before you approve it.

Staleness you can sort by

Every artifact carries its age, so you can clean what you genuinely stopped using instead of the project you opened this morning.

A record of what you freed

Each clean is kept locally with the paths, the commands and the bytes recovered — including the honest note when a command could only be estimated.

In your menu bar or system tray

Free space at a glance and a one-click scan, without leaving whatever you were doing. A native, GPU-rendered app on every platform — no webview, no browser engine.

Yours to configure

Scan roots, enabled ecosystems, staleness thresholds and a blocklist of folders Void must never touch. Settings are plain JSON on disk.

Built for the AI era

Agents work in parallel, and every parallel session leaves a worktree, a transcript and a build folder behind. Void knows where they all live.

Agent worktrees

Finds worktrees from Claude Code, Cursor, Codex and Conductor, with branch, uncommitted and unpushed work, merge state and lock. Trim the build artifacts and keep the worktree, or remove a clean one with git worktree remove — re-checked when it runs, never --force.

Agent data, kept tidy

Old Claude Code transcripts, file history and debug logs, Codex sessions, and the state.vscdb files Cursor, VS Code and Windsurf grow forever — compacted with VACUUM, never deleted, so your chat history stays. Settings, credentials, memory and CLAUDE.md are never touched.

Local models

Ollama per model, counting shared layers once, plus orphaned blobs. The Hugging Face cache per repo, with detached revisions and incomplete downloads. LM Studio, torch hub and ComfyUI too.

Duplicate models, without deleting

The same weights downloaded into two stores are replaced by APFS copy-on-write clones (or hardlinks). Every path keeps working, the space comes back, and contents are re-hashed right before anything changes.

Usage by agent

Everything added up per tool, so you can see that Claude Code is costing you 38 GB — 30 in worktrees, 6 in transcripts, 2 in browsers.

Project graveyard

Git projects with no commit and no edit in 90 days, with their remote, unpushed commits and dirty state up front. Strip the artifacts and keep the code, or move the project to the Trash.

Your agents can call Void

The agent filling your disk is the first to notice it filling up. Give it a CLI, an MCP server and a hook, and it can ask for a cleanup plan itself.

Claude Code
$ claude mcp add void -- void mcp
$ void hook install
# SessionStart: warn when disk is low
# WorktreeRemove: trim build artifacts
Any shell, any agent
$ void scan --json
$ void usage
$ void plan      # dry run, expires in an hour
$ void apply     # only a plan you reviewed
$ void mcp-config # snippet for other MCP clients

MCP tools

  • disk_status free and used space on the volume
  • scan everything reclaimable, by ecosystem
  • explain_item what an item is and what each action would do
  • plan_cleanup a dry-run plan
  • apply_plan runs a plan, only with explicit confirmation
  • usage_by_agent disk usage per AI tool

Danger actions are never offered over MCP. Anything permanent stays with you.

Guard mode

Void watches free space in the background and tells you before a build fails for lack of it.

Warned early

A notification at 15% free and an urgent one at 5%. The menu bar shows where you stand, and Void can launch at login.

Auto-clean, if you want it

Off by default. Turn on a policy — trim idle agent worktrees, clear old agent logs, remove stale build folders — and it runs Safe actions only.

Nothing is deleted without your say-so

Void is a tool that removes files, so it is built to be argued with rather than trusted blindly.

Safe Caution Danger

Risk belongs to the action

An item can offer several cleanups at different risk levels. Void shows the risk of the one that will actually run, and lets you pick a different one.

Friction that scales

An all-safe batch is one click. Anything riskier is itemised with its command. Permanent deletion asks you to type delete.

Places it refuses to go

Keychains, SSH and GPG keys, cloud credentials, shell config, your Documents folder, your home directory itself, and your agents' settings, credentials and memory are protected regardless of how Void is configured.

Git work, checked twice

A worktree is removed only if it is still clean when the action runs. Void never passes --force, and deletes branches with -d, so git refuses anything unmerged.

Sizes you can trust

Hardlinked files are counted once and sparse files by what they really occupy, so the space Void promises is the space you get back.

The Trash, where possible

When an artifact can be moved rather than destroyed, Void prefers that by default — and says plainly which actions cannot be undone.

Fifteen ecosystems

Void knows what each toolchain leaves behind and how that toolchain prefers to clean it up — from uv, Bun and Playwright browsers to Docker build cache and Next.js caches.

Agent worktrees AI agent data Local AI models Stale projects Rust Node.js Python Go Java / JVM Docker Apple / Xcode Homebrew JetBrains .NET System