Skip to main content

Roadmap

This roadmap describes direction, not a compatibility promise. For behavior you can rely on today, use the user guide and the release notes.

Current foundation​

The project already includes:

  • non-overlapping cleanup plans with accurate selected-space totals;
  • system-trash cleanup and manifest-based restore on macOS, Windows, and Freedesktop-compatible Linux desktops;
  • per-item cleanup and restore results;
  • cancellable single-pass scanning, glob ignores, and known-cache discovery;
  • execution-time path, type, file size, directory fingerprint, modification, and protected-path checks;
  • a local authorization boundary that external tools cannot bypass;
  • a versioned, read-only local analysis report for external local agents, alongside scan JSON, plan export, dry-run, and restore commands;
  • versioned, declarative plugin bundles with compatibility and trust metadata.

Near-term: clearer control and recovery​

Planned work includes:

  • clearer retries for partial cleanup and restore failures;
  • large-tree performance benchmarks and broader cross-platform restore tests;
  • more visible access to manifest and diagnostic details from the TUI.

Developer-cache intelligence​

The project intends to deepen developer-specific guidance:

  • broader cache coverage for package managers, build tools, IDEs, mobile toolchains, and containers;
  • scoring that considers safety, reclaimable space, observed modification recency, and rebuild cost; it must not present modification time as proven last use;
  • conservative, balanced, and maximum-space presets;
  • explanations of how each cache is recreated and whether network access is required;
  • validated and signed distribution for community rule packs.

Automation​

Potential automation surfaces include scheduled diagnostics and richer machine-readable failure reports. Any execution surface must remain tied to an explicit, locally reviewed user action.

AI is an external consumer of local evidence and a possible rule-authoring assistant. Cleanr will not embed a model or provider, grant an AI cleanup permission, or turn a suggestion into unattended destructive action. Remote sharing, if ever considered, requires a separately designed redacted contract.