claude-mem-lite plugin
claude-mem-lite keeps decisions, fixes and context from Claude Code sessions in one SQLite file with full-text search
- 68 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Hooks read tool actions and output in every Claude Code session and write summaries to a local database
- Episodes are sent to a small model for summarizing
- Installation compiles a native dependency and edits hook settings
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/claude-mem-liteDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Hooks read tool actions and output in every Claude Code session and write summaries to a local database
- Episodes are sent to a small model for summarizing
- Installation compiles a native dependency and edits hook settings
This is third-party code. Review the repository files before installing.
What it does
claude-mem-lite is long-term memory for Claude Code as an MCP server plus a set of hooks. The hooks automatically capture observations, decisions and bug fixes as you work, and FTS5 search with BM25 ranking and query expansion brings them back in later sessions. Data lives in one SQLite file in ~/.claude-mem-lite/, with no external services. The project is a lightweight redesign of claude-mem. Instead of calling a model for every action it batches 5 to 10 operations into one episode and then sends the summary to a small model, while a deterministic code filter cuts noise. The hooks cover all seven lifecycle events: SessionStart, PreCompact, PreToolUse, PostToolUse, PostToolUseFailure, Stop and UserPromptSubmit. The mem-lite MCP server defines 19 tools, with the core ones listed by default: mem_search, mem_recent, mem_recall, mem_timeline, mem_get, mem_save and three deferred-work tools, mem_defer, mem_defer_list and mem_defer_drop. Slash commands include /mem, /lesson, /claude-mem-lite:bug and /adopt. A marketplace plugin install is complete on its own.
Who it is for. Developers who spend a lot of time in Claude Code and want the agent to remember past bugs, decisions and unfinished work across sessions.
Good fit when
- Before editing a file you want past fixes and context for it through mem_recall
- Work is left for tomorrow, and you want to mark it with mem_defer so the next session shows it in the Deferred Work block
- Memory cost is a worry: episodes are batched instead of a model call per action
- You want to search history in Chinese and English with synonyms
- You want memory without external services, in one SQLite file
Not a fit when
- You do not use Claude Code: the plugin is built around that agent's hooks only
- You need vector-based semantic memory for any LLM application: search here is FTS5 full-text
- Hooks must not write data about your sessions to disk
- You are on Windows without Git Bash or WSL: four hooks need them, although install and the MCP server work
Example request
Find past fixes for auth.ts in memory and show what I deferred to the next sessionLimitations
- The plugin works only with Claude Code, on Linux and macOS, with Windows lacking CI coverage and needing Git Bash or WSL for four hooks.
- Installing compiles native better-sqlite3 with npm install --omit=dev.
- The README calls its savings figures architecture estimates without measurements.
- The plugin moves an old claude-mem database from ~/.claude-mem/ into ~/.claude-mem-lite/ and keeps a backup copy.
- Hidden tools stay callable but do not appear in tools/list.
How to disable. For the plugin run /plugin uninstall claude-mem-lite. For npx or git installs the project has its own uninstall, which with --purge removes data in ~/.claude-mem-lite/.
FAQ
How is claude-mem-lite different from claude-mem and mem0?
It takes the idea from claude-mem but batches 5 to 10 operations into an episode instead of a model call per action and runs on three npm packages. mem0 targets any application through an SDK, while claude-mem-lite is built for Claude Code hooks.
How do I install claude-mem-lite?
Run /plugin marketplace add sdsrss/claude-mem-lite and /plugin install claude-mem-lite. There is also the one-liner npx github:sdsrss/claude-mem-lite, and npx claude-mem-lite for the stable release. You need Node.js and Claude Code.
Where is the data stored?
In one SQLite file in the hidden ~/.claude-mem-lite/ folder. There are no external services or vector databases.
How do I mark work for the next session?
Call the mem_defer tool. The deferred item appears in the Deferred Work block at the next session start, mem_defer_list shows open items and mem_defer_drop removes one with a reason.
How do I remove claude-mem-lite?
For the plugin run /plugin uninstall claude-mem-lite. For npx or git installs the project has an uninstall, and the --purge flag also erases data in ~/.claude-mem-lite/.
Discussion
No questions or reviews yet.
Related
PluginsThe Supermemory plugin for Claude Code: persistent memory across sessions with automatic prompt recall, conversation capture and a codebase index
ClawMem gives Claude Code, OpenClaw and Hermes local memory: note search, session hooks and a 33-tool MCP server with no API keys
A persistent memory plugin: stores observations from agent sessions and injects relevant context into new ones
An MCP server that preserves Claude Code context between sessions: decisions, checkpoints and a file cache kept in local SQLite.
A long-term memory plugin that keeps decisions, preferences and task state across sessions and injects only what a query needs
Long-term memory for Claude Code, Cursor and Codex: the agent saves decisions and finds them again in later sessions
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| claude-mem-lite (this card) | no | 68 | |
| Claude-Mem | no | 116,457 | the original claude-mem calls a model for every action, while claude-mem-lite batches operations into episodes and stores everything in one SQLite file |
| Mem0 | no | 68,887 | mem0 serves any LLM application through an SDK, while claude-mem-lite works through hooks and MCP in Claude Code only |
README in short
The README compares the project with claude-mem on architecture and cost, covers three install methods, what the installer does, the MCP tool list with CLI equivalents for hidden tools, search quality and platform support. Version notes and an FAQ come last.