memdebug Audits and Restores AI Agent Memory Locally

Summary

memdebug is an alpha, local-first tool for inspecting and restoring memory used by AI agents. It can work with notes in folders or Git repositories, Open WebUI memory, and self-hosted Mem0. The tool records memory contents in a tamper-evident ledger, shows when snapshots changed, detects edits that bypass a store’s own history, compares snapshots, and flags wording for review. It can safely roll back Markdown memory, with a dry-run mode and an explicit apply step. memdebug operates as an observer: it never sits between an agent and its memory, never communicates with the agent, and runs entirely on the user’s computer. It therefore does not block attacks in real time and should be used alongside runtime guards. For Claude Code, it can link a flagged change to the logged session as evidence, but it cannot identify which conversation wrote memory for other agents. The project is at alpha status, and its tests run on Linux, Windows, and macOS across multiple Python versions. It requires Git 2.31 or newer and Python 3.10 or newer, with standalone binaries also available; the ledger should be treated as sensitive because it contains agent memory text.