AI memory for agents

Your notes know more. Put them to work.

Keep the project rules, discoveries and decisions Claude Code already wrote down. We bring those notes into persistent, searchable AI memory, so you can find the knowledge you need for the next task. The import preserves the prose without another observation-generation pass.

Claude-Mem mascot carrying a useful project note

AI memory in production

70 days of work. One searchable memory.

Alessandro used our open-source engine across three machines for 70 days. His customized deployment captured the decisions, fixes and discoveries behind the work—and published the data.

Observations captured
22,718
Bug fixes recorded
1,262
Decisions recorded
757

See how he made that knowledge searchable. Then give your own agents a memory to build on.

Claude-Mem’s mascot files a note in a shared card catalogue connected to three computers.
Alessandro’s published field study · March–May 2026

Reviewed November 18, 2026 · By Claude-Mem

01

Give existing knowledge a searchable home.

The source is the memory directory inside Claude Code’s projects directory. Topic Markdown files contain the knowledge; the link-only MEMORY.md index is skipped. This is a note import, not a transcript import or a general-purpose folder crawler.

02

See what you will bring across.

Run the dry-run command from the project you want to import. It scans and counts without starting the worker or writing the database. Review the selected source and any skipped or failed files before choosing the real import. Use a current installation that exposes memory ingest; update if the command is unavailable.

03

Keep the knowledge. Skip the extra generation.

The real import starts the worker if needed and stores the parsed topic prose through the observation path. Ingest itself does not call the observation-generation model. This does not make unrelated provider activity or later model-assisted recall free.

04

Keep the next revision, too.

Content-hash deduplication reports unchanged notes as already imported. Changed prose can be stored as a new observation, even with the same title, and earlier revisions remain. The result summary separates stored, already-imported, skipped and failed files.

05

Respect the source boundary.

An explicit source must remain inside Claude Code’s projects directory, including a configured CLAUDE_CONFIG_DIR. Symlinked notes and files over 64 KB are skipped. Frontmatter-only files are empty; supported metadata is preserved. A sibling transcript can supply project identity, but its conversation content is not ingested.

06

Find the rule when the next task needs it.

Search for a distinctive phrase from one topic note, open the imported observation and check project identity. If the scan cannot resolve a project’s working directory, inspect its fallback grouping; the documented --require-cwd option skips unresolved project directories. Keep the original notes as your source.

Start with the supported commands

Preview notes for the current project

npx claude-mem memory ingest --dry-run

Import after reviewing the scan

npx claude-mem memory ingest

Optional: preview every project before a broader import

npx claude-mem memory ingest --all --dry-run

Questions before you start

Does ingest import my full Claude Code conversations?

No. It imports the distilled Markdown topic notes. It may read a sibling transcript for its working-directory identity, but does not ingest that transcript’s conversation content.

Can I point it at any Markdown folder?

No. The documented source boundary is Claude Code’s projects directory. Sources outside it are refused, and symlinked notes are not followed.

Does it replace or rewrite the original notes?

The ingest workflow reads topic files and stores observations in Claude-Mem. Keep those original files; this is an import into searchable history, not a bidirectional note editor.

Can I carry that knowledge to my other tools?

Connect a supported agent to Claude-Mem and verify that it can recall the imported observation in the intended project and host scope. CMEM Pro adds eligible-record sync across connected devices; neither import nor sync copies every local file.

Go deeper

Explore the source documentation and continue with the topic that matches your work.

Give your next task a stronger starting point.

Install Claude-Mem, preview your notes and make the useful knowledge searchable. Add CMEM Pro when you want eligible memory synced across connected devices.