Remember.Plugin

Personal memory system for Cowork and Claude Code

A plugin for Claude Cowork and Claude Code that saves your conversations as you chat and turns them into structured markdown notes.

Download remember.plugin View on GitHub

Open any Claude Cowork conversation. Type /remember and your chat gets distilled into typed markdown files — decisions, preferences, facts, people — saved into a folder you control.

As you keep talking, memory updates continuously in the background. Bring in existing documents with /ingest. Run /lint every few weeks to keep things tidy.

Zero infrastructure. Pure markdown. Your data stays on your machine.

Install

Download remember.plugin using the button above
Open a new Cowork chat and drag the downloaded file into the chat input
Type: Show this plugin file as a card so I can save it.
A card appears with a Save plugin button — click it
Go to any Claude Code / Cowork chat and type /remember to get started

What you get

Skill	What it does
Remember	Capture durable facts from the current chat into typed atomic files + daily journal
Ingest	Bring existing documents (MD, TXT, PDF, Word, PowerPoint) into memory. Only new and changed files processed by default.
Start memory folder	When you don't have a folder yet, sets up one for you
Bootstrap	Set up the memory structure in any folder; chains to Ingest if documents are found
Lint	Layered health check across atomic memory, journal, CLAUDE.md, and wiki. Auto-fixes legacy patterns.

What it does for you

Notes that organize themselves. Your conversations become markdown files — decisions, references, people, preferences — automatically sorted into the right folders.
Works while you talk. Important facts get captured in the background as you chat. No commands needed for everyday capture, though you can always use /remember to checkpoint explicitly.
Bring in existing documents. Drop PDFs, Word docs, presentations, or markdown into your folder. The plugin reads them, extracts the durable facts, and adds them to your memory. Add more later — only the new ones get processed.
Per-project, not one big brain. Each project has its own memory. Share a project's folder for work without exposing your personal notes. Archive a project without entangling others.
Auto-captures people. When you mention someone with real context (a role, a decision attributed, an action they took), a profile gets created or updated. No more "wait, who was that person again?"
Works with Obsidian. Open your folder as an Obsidian vault for a graph view of how everything connects. The plugin uses Obsidian-compatible markdown.
Files you own forever. Pure markdown in folders you choose. Open in any text editor. MIT licensed and open source. Uninstall the plugin and your notes keep working.
Educational chaining. When the plugin does something automatically, it tells you the underlying command so you can use it directly next time. The plugin teaches itself through use.

Architecture

Dual-layer model: atomic memory (typed facts) + synthesized wiki (narrative pages with cross-links). Atomic files live at memory/{feedback,projects,reference,people}/. Wiki pages use [[wikilinks]] — open your folder as an Obsidian vault if you like, it works natively.

Inspired by Andrej Karpathy's LLM-Wiki insight (local markdown as the LLM's context substrate), structured using Andy Matuschak's evergreen + structure notes pattern, with Niklas Luhmann's Strukturzettel as the wiki layer.

Privacy & data

Everything stays local. Files are created in folders you choose. Zero network calls beyond Claude's normal conversation traffic.
No telemetry. The plugin doesn't phone home.
No accounts. No signup. No login. No vendor lock-in.
Open source. Read the code, fork it, modify it.
Portable. If you uninstall, every file the plugin created is still readable in any text editor.

Compatibility

Claude Cowork (Anthropic's desktop knowledge-work app)
Claude Code (CLI — same plugin format)
Files are pure markdown — readable in Obsidian, Logseq, any text editor