Claude Recall vs. the official Anthropic memory tool
Anthropic ships a memory tool with Claude Code. Multiple users have reported memory files vanishing between sessions, consolidated in the canonical thread GitHub #36091 (with duplicates filed under #38459, #25032, and #33912). This page is the honest version of “should you use it, us, or both.”
TL;DR
- Use the official tool if you have fewer than thirty sessions and have never lost a file.
- Use Claude Recall if you have fifty-plus sessions and care about durability.
- Use both if you want the official tool's tight Claude Code integration plus a verifiable on-disk backup.
Side-by-side
| Capability | Anthropic memory | Claude Recall |
|---|---|---|
| Ships in Claude Code | Yes | Independent install |
| Survives session restart | Open bug #36091 + dupes | Yes |
| Three-layer durability | No | Yes |
| Searchable across sessions | Limited | Yes (FTS5) |
| Local SQLite database you can grep | No | Yes |
| Plain-text disk mirror | No | Yes |
| MCP server | No | Yes |
| Web UI / browser timeline | No | Yes |
| Cost analytics | No | Yes |
| Built into Claude Code surface | Yes | Separate install |
Where they win
The official memory tool is built into Claude Code. There is nothing to install, nothing to learn. If you are running fewer than thirty Claude Code sessions and have never lost a file, that built-in surface is the right answer. We respect it. The reason this page exists is the open canonical bug at #36091 (and several duplicates), where memory files disappear between sessions; for any user with real history, that bug is the line.
Where we win
Claude Recall guarantees three-layer durability. Every write hits SQLite (with a history column for full versioning), then a plain-text mirror on disk that is grep-able from any tool. The source JSONL files Claude Code already writes are never modified. This is engineered into the schema, not promised in a blog post; the migrations are inspectable in the project. The official memory tool currently makes no such guarantee. Bug #36091 (and its duplicates) is the test: a recurring user-reported file loss that exists in the official tool and cannot exist in Claude Recall by design.
When to pick which
- Use the official memory tool when you are new to Claude Code and your usage is small.
- Use Claude Recall when you have history that matters.
Coexistence
They do not conflict. Claude Recall reads the JSONLs Claude Code already writes. The official memory tool can run alongside without interference. If you ever decide to switch, your data is yours either way; nothing leaves your machine in either case.
FAQ
What if Anthropic fixes #36091?
Claude Recall is still the only tool that gives you searchable cross-session memory, a plain-text mirror, an MCP server, cost analytics, and a web UI. The bug is the most-citable instance of the difference, not the only one.
Will Anthropic ship native session search?
Possibly. We watch for it. If they do, Claude Recall pivots to the parts they will not build (cross-tool memory, structured exports, team sharing).
Is this disrespectful to Anthropic?
No. Claude Recall is built for Claude Code users; we want Claude Code to win. Pointing to a known issue with a verifiable fix path is the most respectful thing we can do.