Keep your AI's memory tidy, honest and conflict-free.
A deterministic command-line tool that audits and consolidates Claude's file-based memory across many areas — finding duplicates, stale facts, broken links and conflicting versions, then helping you resolve them safely. No model inference, no network calls: the same input always gives the same result.
Linux, Windows & macOS · ~2,000 lines · 194 tests · MIT licence
Everything runs locally and deterministically — nothing is guessed by a language model.
Per-area checks for spec conformance, duplicates, staleness and broken links — a health report for each memory area.
Finds slug conflicts and near-duplicates across every area — the signal that two areas have drifted apart.
Pick a canonical version for diverged content; the tool writes it once and turns the originals into pointer stubs.
Preview safe fixes on a disposable copy and diff them before anything real is ever touched.
Human-approved choices are committed to git in review_decisions.json, so every consolidation is auditable.
Discover memory-shaped files outside your configured areas, and initialise a fresh machine from scratch.
Memory Consolidator is MIT-licensed and lives on GitHub, with continuous integration across Linux, Windows and macOS. Clone it, read the tests, and run audit against your own memory. Get it on GitHub →