AI memory for agents

Keep the knowledge. Choose what stays.

Your agents need the lessons, decisions and discoveries that move work forward. We give you controls to keep that useful context while excluding selected text from supported memory capture. Use private tags deliberately, and add optional pattern redaction for recognizable secrets.

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

Keep the lesson beyond the session.

Use a paired lowercase <private>…</private> block for text you want removed from supported Claude-Mem capture paths. The current conversation can still contain that text. These tags do not change what your host or its model provider received.

02

Make your choice explicit.

The tag filter removes balanced blocks. An unclosed tag remains literal text, so do not rely on a missing closing tag to exclude the rest of a message. Try the exact paired syntax with a harmless synthetic marker before using it in your workflow.

03

Add a filter for accidental secret output.

Auto-redaction is off by default. Enable CLAUDE_MEM_REDACT_ENABLED to replace matches from the built-in secret patterns. Generic password and token patterns require the separate CLAUDE_MEM_REDACT_GENERIC_ENABLED setting. Detection is based on configured patterns, not an assurance that every sensitive value is found.

04

Preserve the context around a match.

Private blocks are removed before optional redaction scans the remaining text. Redaction replaces matched content with a typed placeholder and preserves surrounding context. It is destructive replacement, not encryption, and enabling it does not erase records already stored.

05

Verify the memory you meant to keep.

Inspect a synthetic example through your actual client and capture path, then check saved prompts and observations. Check your actual integration and host rather than assuming identical coverage everywhere. Keep real credentials in your normal secret store.

06

Review connected services separately.

Local memory controls and Cloud Sync are different choices. Cloud Sync can upload eligible saved memory, and optional setup analysis has its own selected-history submission. Check the active provider and sharing setup; private tags are not a promise that an entire workflow stays on one machine.

Questions before you start

Are private tags an encryption feature?

No. They remove matching blocks from supported memory capture. They do not encrypt your host transcript or already-stored data.

Will auto-redaction catch every secret?

No. It matches configured patterns and can miss values outside those patterns. Keep credentials out of prompts and tool output where possible, and test your own capture path with synthetic examples.

Does redaction cover server-runtime metadata?

Not uniformly. The current server runtime redacts event-body text, but session and event metadata use a tag-dependent path: untagged strings do not receive pattern redaction there. Keep credentials out of metadata and verify your capture path with synthetic values.

Is auto-redaction enabled automatically?

No. Built-in redaction and additional generic patterns are separate opt-in settings. Follow the current configuration guide and verify the saved result.

Go deeper

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

Build memory worth carrying forward.

Install Claude-Mem to capture useful knowledge with deliberate memory controls. Add CMEM Pro when you want eligible memory synced across connected devices.