Claude-Mem: Persistent Memory, Setup Sources and Privacy Checks

Claude-Mem is a persistent-context system for Claude Code. Its documentation describes capturing tool observations, summarizing them and making context available in later sessions. The linked source project is thedotmack/claude-mem.

What memory changes

A new session may need facts from earlier work. Claude-Mem's documented approach stores observations and summaries so they can be retrieved or injected later. This is different from assuming that a fresh model session remembers every earlier message accurately.

The documentation lists installation routes, restart instructions, search tools, a viewer, configuration and privacy controls. It includes manual private tags and optional redaction. Sources checked October 30, 2026. Such controls should be reviewed in the configuration you use; the feature list alone does not establish that every secret is excluded.

Set it up deliberately

Use the official installation documentation for your current client and version. Before enabling it on sensitive work, establish what is captured, where it is stored, which model/provider processes it, and whether any hosted or synchronization option is enabled.

Start on a harmless project. Record a simple project decision, end the session, then ask a new session to retrieve it with the supporting observation. Check that the returned context matches the original rather than merely sounding plausible.

Memory versus project instructions

Treat retrieved memory as evidence to check. Old context can become stale, and a compressed summary may omit the condition that originally made a decision valid. Preserve source references for decisions that matter.

Varwo has not installed the plugin, inspected a live memory database or tested its redaction. This page explains the documented workflow and a suggested local check; it does not claim measured recall accuracy or privacy guarantees.