AI memory for agents

Bring the knowledge. Build from here.

Your agents have already learned a lot. We bring that knowledge into persistent, searchable AI memory: import from Mem0 or Supermemory, recover Claude Code notes, or move your existing Claude-Mem records. Keep the decisions and discoveries that make the next session stronger.

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

Start with the memory you have

Four paths. One stronger starting point.

Choose your source. We keep setup and verification specific to the memory you are moving.

From Mem0

Bring the memories for one Mem0 user into searchable observations. Preview the selected scope before you write.

Import Mem0 memory

From Supermemory

Bring current extracted memories from one namespace. Or choose published document text with the separate documents mode.

Import Supermemory memory

From Claude Code notes

Turn existing Markdown topic notes into searchable observations, without another observation-generation pass.

Recover Claude Code notes

From another Claude-Mem installation

Export selected native records, review the JSON and import it with the destination worker.

Move native memory

For Mem0 and Supermemory, use Claude-Mem 16.0.0 or a newer compatible release and a source API key. A dry run fetches and validates your selected scope without writing; the real import needs a running local worker. Read preview and verification steps.

Reviewed November 18, 2026 · By Claude-Mem

Move native Claude-Mem records

01

Choose the history you want to move.

The native export searches your saved history and writes a JSON file containing matching observations, session metadata, summaries and prompts. Use a project filter for a focused transfer. An empty query selects all history within the chosen project; it is broader than a normal topical search.

02

Run the scripts from a Claude-Mem checkout.

These are repository scripts, not commands available from any directory. Use a current Claude-Mem source checkout with its dependencies installed and a running source worker. Export there; on the destination, use its checkout and running worker to import the file. The documentation covers setup and current requirements.

03

Review the file before it leaves your machine.

The JSON contains readable memory content, including matching prompts and potentially private paths or other sensitive material. Keep the file in a destination you control, inspect what it includes and share only the records you intend to move.

04

Read imported, skipped and rejected separately.

The importer sends records to the destination worker. Already-present records can be skipped; invalid or refused records can be rejected. Review the printed counts and reasons. A nonzero exit status or rejected count needs investigation even if other records were imported.

05

Verify a known memory at the destination.

Search for a distinctive decision from the export, inspect its observation and surrounding session, and confirm that it belongs to the expected project. Keep the source installation and export until the destination contains the records you need. A successful command alone is not a content check.

06

Choose transfer, note ingest or ongoing sync.

JSON export/import moves native memory records. Claude Code note ingest imports its distilled Markdown topic files. Cloud Sync handles ongoing eligible-record updates between connected devices. These paths have different scope; none is a promise to copy every local file or restore every application setting.

Start with the supported commands

Source checkout — choose your actual project name

npx tsx scripts/export-memories.ts "" project-memory.json --project=my-app

Destination checkout — running destination worker required

npx tsx scripts/import-memories.ts project-memory.json

Questions before you start

Does this export every local file?

No. The native memory export covers matching observations, session metadata, summaries and prompts. Work-state exports use a separate documented endpoint; they are not a work-state import format. Arbitrary project files and media are not promised by this JSON workflow.

Will running import twice duplicate everything?

The native importer detects already-present records and reports them as skipped. Check the results for rejected rows and verify important records; duplicate prevention is not a guarantee that every possible export has imported completely.

Can I import from Mem0 or Supermemory?

Yes. Our direct-provider command reads one Mem0 user or Supermemory namespace into searchable observations. Use Claude-Mem 16.0.0 or a newer compatible release, a source API key and a stable source-account label. Preview with --dry-run, then start the local worker before the real import. The guide explains exact commands and result checks.

Does a provider import keep syncing with the original service?

No. Provider imports read your selected source without changing or deleting it. They are one-way imports, not ongoing synchronization or graph migrations. Source deletions and future expirations are not synchronized after import; review those records before relying on them.

What does Supermemory document mode bring across?

The default imports current extracted memories. The optional --documents mode imports published extracted text, not the original uploaded files. Choose the mode you need; importing both can create overlapping information.

How do I use imported memory in another agent?

Provider imports file observations under the default Claude host. Use the documented --platform option to choose another host, such as Codex or Cursor. Then connect your supported agent and verify recall. Ongoing sync across connected devices is a separate CMEM Pro feature.

Go deeper

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

Make what you know your next advantage.

We turn retained knowledge into a stronger next session. Install the memory engine, bring your context, and add CMEM Pro when you want eligible memory synced across connected devices.