compound-memory plugin

compound-memory turns your AI chats into memory about you each night and injects it into new Claude Code and Codex sessions

  • 113 ★ on the repository
  • Last commit
  • MIT
Plugin

Broad access

Why this level

  • Reads every Claude Code and Codex session log on the machine and sends them to a model for distillation
  • Edits `~/.claude/settings.json` and `~/.codex/AGENTS.md` and installs a nightly cron job
All reasons and checks

protocol-zero-0/compound-memory

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add plugins/compound-memory

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • Reads every Claude Code and Codex session log on the machine and sends them to a model for distillation
  • Edits `~/.claude/settings.json` and `~/.codex/AGENTS.md` and installs a nightly cron job

This is third-party code. Review the repository files before installing.

What it does

A background process reads the session logs of Claude Code, Codex and DeepSeek Harness once a day, extracts what concerns you and writes a "current state" snapshot. At the start of a new session the snapshot is placed into context automatically. Memory splits into a long-lived part, such as requirements you stated, and a short one that holds unfinished business and promises. `cm recall` searches memory and names the source of each entry. Stale entries lose force when you change your mind. Original chats stay on the machine, and only entries judged necessary and fit for sharing leave it, with keys stripped first. By default up to 20 sessions are processed per night. `cm distill --dry` shows the volume in advance. Installation runs `install.sh`, and the Claude Code plugin adds a `cm recall` call on the agent's request.

Who it is for. Developers who work with Claude Code and Codex on several machines every day and are tired of explaining their habits and agreements to the agent again.

Good fit when

  • Every new chat starts with a story about you and the project
  • You work in Claude Code and Codex in turns and want one memory for both
  • You must quickly recall what was agreed with a supplier and see which session the fact came from
  • You want a check before the nightly run: how many sessions go to the model
  • The install must be rolled back fully, including the entries in `settings.json` and `AGENTS.md`

Not a fit when

  • You are not ready to hand all session logs to a model: they are the only source of data
  • You need memory about a project: the tool builds a user profile
  • The model budget is limited: the nightly run uses a model, Claude Opus by default

Example request

Install compound-memory, run cm distill --dry and show how many sessions would be processed tonight

Limitations

The tool reads all your session logs and has no other source. Each night it calls a model: Claude Opus by default, and on Sonnet the author measured about a quarter of entries going missing. Installation edits `~/.claude/settings.json` and `~/.codex/AGENTS.md`, a backup is made first and `cm uninstall` removes the markers. A working model call is required: a local `claude` command or an OpenAI-compatible endpoint. The main description is in Chinese, with an English version in `README_EN.md`.

How to disable. Run `cm uninstall`: it removes the entries from settings and cron. With `--purge` the accumulated memory entries are deleted too.

FAQ

How do I install compound-memory?

Clone the repository and run `bash install.sh` inside it. The installer runs the first distillation and prints the snapshot. A Claude Code plugin from the repository marketplace also exists.

What does the tool read?

Session logs of Claude Code, Codex and DeepSeek Harness on your machine. `cm distill --dry` shows the volume before the first model call.

Do my chats leave the machine?

Originals stay local. Only entries judged necessary and fit for sharing leave it, with keys stripped first.

How do I find a fact in memory?

Run `cm recall "supplier"`: the result shows the entry, the date and the session source.

How do I remove it?

`cm uninstall` restores the settings files, and `cm uninstall --purge` deletes the memory too.

Discussion

No questions or reviews yet.

Plugins

Local memory for Claude Code: a single binary indexes past conversations and injects relevant context into new sessions.

MCP serverBroad access230Repository stars

A SQLite archive of Claude Code sessions with an MCP server, a terminal UI and a web UI for finding past conversations.

CLIExtended access64Repository stars

A Claude Code plugin that keeps memory across sessions in one local file, with no database, cloud or API keys

PluginExtended access586Repository stars

A skill for searching past sessions: a full-text index of Claude Code, Codex, pi and Grok chats in one SQLite database

SkillMinimal access146Repository stars

Eight Claude Code skills for Obsidian: Granola notes, NotebookLM imports, TaskNotes tasks, recall of past sessions, chat export

PluginExtended access552Repository stars

A plugin that turns Claude Code into a personal chief of staff with persistent memory, a morning brief and a weekly review

PluginExtended access92Repository stars

Comparison with similar tools

Cards that share the main search query, and how they differ.

ToolOfficialRepository starsMain difference
compound-memory (this card)no113
Claude-Memno116,457compresses sessions into project context, while compound-memory builds memory about the user shared across three tools
Episodic Memoryno499searches old chats, while this one runs a nightly distillation that retires stale entries

README in short

The Chinese README explains the idea: a nightly process turns your conversations with AI into memory that is injected at the start of the next chat. It shows a sample snapshot and recall, lists warnings before install, the install and removal commands, model requirements and links to design, recall and privacy gate documents.

README badge

Report a problem