v0.4.0

Behaviour changes — read before upgrading

• Flatpak app caches are cleaned by default. ~/.var/app/<id>/cache used to need --flatpak because reclaim could not tell which flatpak apps were running. It can now, so a running app's cache is skipped and reported, like any other app's. --flatpak now covers only removing unused runtimes.
• npm cache clean and uv cache clean moved from tier 0 to tier 1. New tier-0 units run first: npm cache verify and uv cache prune drop only entries nothing uses. A plain clean --apply still runs both.
• Under --free and --auto, some caches are trimmed instead of wiped. The Go build cache, pip, npm's _cacache and thumbnails give up their least recently used files first, and the run stops once the target is met. Without a target, every cache goes whole as before.
• --models no longer deletes your Hugging Face login. It used to remove all of ~/.cache/huggingface, including token and stored_tokens. It now takes only hub, xet, datasets and assets.
• Sizes are now allocated disk space, not file length. Sparse files, hard links and cloned files no longer inflate the numbers. Space that is hard-linked from outside a cache is shown as "shared, not freed".

New

• macOS binaries. This is the first release with darwin/amd64 and darwin/arm64 tarballs, plus .icns. They are unsigned; see the README for the xattr step.
• macOS --system: the unified log (log erase --ttl) and Time Machine local snapshots (--timemachine, lossy). Also added: macOS idle-app detection, Homebrew Cask and Steam units.
• Ollama models (--models --allow-lossy) are removed through ollama rm, naming each model the server lists. reclaim never deletes blobs under a live store.
• Chromium and Electron apps: caches in every profile and in Partitions/* (Postman, Slack, VS Code and others). --site-data offers web apps' offline storage as lossy. An app holding SingletonLock keeps its caches.
• Relocated caches are found. reclaim follows XDG_CACHE_HOME, HF_HOME, HF_HUB_CACHE and TORCH_HOME. A variable aimed at / or at your home folder is refused.
• Idle projects: a vendor/ or dist/ directory is claimed only if git proves it is untracked and ignored. Directories marked with CACHEDIR.TAG are claimed too.

Safer deletion

• Deletion now opens each directory relative to its parent and checks it is still the same one, so a directory swapped for a symlink mid-run is refused. It also never crosses into another filesystem or a bind mount.
• Read-only trees such as Go's module cache are removed. Only directories you own are made writable, and their mode is restored if they stay.
• Entries left in place on purpose (mount points, directories refilled mid-run) are listed under Skipped instead of failing the run.
• Command units report the measured change in free space, marked with *, rather than an estimate.
• The test suite runs in a sealed environment. It proved the old tests could reach real caches.

What's Changed

• fix: remove read-only trees such as Go's module cache by @mralaminahamed in #48
• Sealed test harness, anchored removal, honest sizes by @mralaminahamed in #49
• Idle-project scan: nested projects, git proof, CACHEDIR.TAG by @mralaminahamed in #50
• Chromium/Electron caches: profiles, partitions, site data, SingletonLock by @mralaminahamed in #51
• Trim before wipe, and oldest files first under --free by @mralaminahamed in #52
• Cache and model paths follow XDG_CACHE_HOME, HF_HOME, HF_HUB_CACHE, TORCH_HOME by @mralaminahamed in #53
• fix: --models keeps the Hugging Face login by @mralaminahamed in #55
• ollama models, removed through ollama rm by @mralaminahamed in #56
• Running flatpak apps park their own cache by @mralaminahamed in #57

Full Changelog: v0.3.0...v0.4.0

v0.3.0

Behaviour changes — read before upgrading

• The trash is no longer emptied by default. It used to run on every clean --apply as though it were a cache. It now needs --trash --allow-lossy.
• --docker no longer removes stopped containers. A stopped container can hold state that exists nowhere else. They have their own flag, --docker-containers; the dry run names each container, and the apply removes exactly those ids.

New

• reclaim index snapshots every filesystem (616k directories in 16s here; a refresh took 2.4s). analyze answers from an index under a day old: 0.3s where walking home took 2m41s. index show [DIR] and index cold [DIR] query it. It is never used for deleting.
• analyze --apps lists desktop apps unused for 90+ days with the command that would remove each. Advisory only, and only on evidence (binary atime, snap/flatpak data dirs, GNOME Shell's record).
• --obsolete (lossy): config left by removed-but-not-purged packages, rotated logs over 30 days old, directories of superseded JetBrains IDE versions.
• --kernels also takes what removed kernels leave in /lib/modules, only for a version nothing still claims.
• New units: Chrome's on-device AI model (--models), ~/.wp-cli/cache, ~/.yarn/berry/metadata.
• macOS: safe to run, and useful for developer caches (Xcode, Homebrew, CocoaPods, SwiftPM). No darwin binaries yet; see #5.

Fixed

• The dry run reported 0B for apt-get clean (it drops the package indexes too) and for the journal vacuum.
• Under sudo, the index and the operations log went to /root; they now go to the invoking user and are handed back.
• --discover could delete reclaim's own index.
• The "Locked" section ignored --only, --exclude and missing flags.

What's Changed

• fix: status printed nothing at all inside a container by @mralaminahamed in #39
• feat: macOS support for the parts that can be got right by @mralaminahamed in #40
• feat: obsolete files, kernel residue, honest system sizes by @mralaminahamed in #42
• feat: Chrome on-device model, wp-cli and yarn metadata, old JetBrains dirs by @mralaminahamed in #43
• feat: analyze --apps reports applications unused for 90+ days by @mralaminahamed in #44
• feat/index by @mralaminahamed in #45
• fix: four bugs found by a real full cleanup run by @mralaminahamed in #46
• fix: under sudo, the operations log goes to the invoking user by @mralaminahamed in #47

Full Changelog: v0.2.1...v0.3.0

v0.2.1

What's Changed

• fix: the release published no .rpm, and would have published an unfindable one by @mralaminahamed in #38

Full Changelog: v0.2.0...v0.2.1

v0.2.0

What's Changed

• fix: re-rate discovered caches by size, and offer what is held back by @mralaminahamed in #21
• feat: load unit definitions from files by @mralaminahamed in #22
• feat: reclaim unused flatpak runtimes and app caches by @mralaminahamed in #24
• feat: purge superseded kernels, and let a command report what it frees by @mralaminahamed in #25
• feat: reclaim old crash dumps, bounded by age by @mralaminahamed in #26
• feat: name the model caches, and prune the dead half for free by @mralaminahamed in #29
• feat: teach scan the rest of the common project shapes by @mralaminahamed in #30
• feat: shell completion for bash, zsh and fish by @mralaminahamed in #31
• feat: read defaults from a config file, but only ones that narrow by @mralaminahamed in #32
• feat: --below, and systemd units, instead of a watch daemon by @mralaminahamed in #33
• feat: record what was removed, not just which unit ran by @mralaminahamed in #34
• feat: report stale installers, and never delete them by @mralaminahamed in #35
• feat: support every Linux distribution, and install like one by @mralaminahamed in #36
• docs: reorganise the README around what the tool knows, not when it was added by @mralaminahamed in #37

Full Changelog: v0.1.2...v0.2.0

v0.1.2 — honest failure reporting for --system

Running --system used to print warning: system-apt: exit status 1, then list the failed units under Reclaimable as though they had freed space. One of the three failed completely silently.

• Root-requiring units declare it, and elevation is acquired once per run via sudo -v with the terminal attached — one visible password prompt instead of three invisible failures. Dry runs never ask.
• A command's own output reaches its error, so you see sudo: a password is required rather than exit status 1.
• Failed units get their own report section and are excluded from the freed total, in text and JSON.
• The snap pipeline no longer exits 0 when every removal inside it failed.

Also untracks a 5 MB binary committed by accident in 62c32a8.

108 tests across 13 packages, race-clean.