Nemp Memory

Local project memory for Claude Code and OpenClaw.

Save a decision once. Every new session starts knowing your stack, your rules and why things are the way they are.

Why Nemp

Claude Code starts every session from zero. You end up re-explaining the same things:

• your stack and project structure
• decisions you already made, and why
• conventions, preferences and known gotchas
• what you were working on yesterday

Nemp keeps that context in plain JSON inside your project, lets you search it, and keeps your CLAUDE.md in step with it. No servers, no database, no API keys, no dependencies.

Quick start

# 1. Install
/plugin marketplace add https://github.com/SukinShetty/Nemp-memory
/plugin install nemp

# 2. Let Nemp learn your stack
/nemp:init

# 3. Save something worth remembering
/nemp:save api-style "REST, not GraphQL - team decision Jan 2026"

# 4. Keep CLAUDE.md up to date automatically
/nemp:auto-sync on

That's it. Next session, find anything with /nemp:context <topic>.

Windows or marketplace trouble? See other install methods and Troubleshooting.

What you get

1. One-command project setup

/nemp:init reads your package.json and saves your framework, language, database and ORM, auth, styling and package manager as memories in one go.

Framework:       Next.js 14 (App Router)
Language:        TypeScript (strict)
Database:        PostgreSQL via Prisma
Auth:            NextAuth.js
Styling:         Tailwind CSS
Package manager: npm

Saved 6 memories.

2. Compact, typed memories

/nemp:save <key> <value> stores a memory with:

• Compression - filler is stripped and values kept short, with technical terms, paths and versions left exactly as written.
• A type - fact, rule, decision, preference, goal, warning and more, inferred from the key and value (or set with --type).
• Attribution - which agent wrote it (main, nemp-init, backend...), so multi-agent projects stay traceable.
• A conflict hint - if a related key already says something different, Nemp warns you when you save.

3. Context search that expands your query

/nemp:context auth doesn't just look for "auth". It expands to related terms - authentication, login, session, jwt, oauth, token, nextauth, clerk - and searches keys and values across project and global memory.

FOUND 3 MEMORIES MATCHING "auth"

auth-provider    NextAuth.js with JWT strategy
auth-tokens      15min access tokens, 7-day refresh
auth-middleware  Protects /api routes except /auth/*

Search is keyword-based with a built-in synonym map. It runs locally and needs no embeddings or model downloads.

4. CLAUDE.md that stays current

Your own rules at the top of CLAUDE.md are never touched. Nemp only manages its own section.

5. Suggestions from your work

/nemp:suggest reads your activity log (files you keep editing, packages you install, commands you repeat) and drafts memories for you to save, edit or skip. Nothing is saved without your say-so unless you use --auto.

Activity capture is opt-in with /nemp:auto-capture on and is still experimental.

6. Audit trail and health check

• Every read, write and delete is logged to .nemp/access.log. View it with /nemp:log, filter by agent, or tail recent entries.
• /nemp:health checks the store: valid JSON, empty or oversized values, duplicate keys, stale CLAUDE.md or MEMORY.md, and gives a score out of 80.

7. Global memory

Preferences that follow you across projects - "prefers Bun", "always use strict TypeScript" - go in ~/.nemp/ with /nemp:save-global. Project memory wins when both have an answer.

Commands

How it works

Nemp is a set of Claude Code slash commands and one skill. When you run a command, Claude follows its instructions to read and write small files in your project. There is no background process, server or database.

.nemp/
  memories.json   # project memories
  access.log      # read / write / delete audit trail
  config.json     # settings such as auto-sync
  MEMORY.md       # human-readable index of memories

~/.nemp/
  memories.json   # global memories

A memory is plain, readable JSON:

{
  "key": "auth-provider",
  "value": "NextAuth.js with JWT",
  "type": "decision",
  "tags": ["auth"],
  "agent_id": "nemp-init",
  "created": "2026-01-31T12:00:00Z",
  "updated": "2026-02-11T14:00:00Z"
}

You own the files. Commit them to share context with your team, or add .nemp/ to .gitignore to keep it personal. Delete everything with rm -rf .nemp.

Nemp vs. CLAUDE.md alone

Installation

Plugin marketplace (recommended)

/plugin marketplace add https://github.com/SukinShetty/Nemp-memory
/plugin install nemp

Windows, if the marketplace command fails

/plugin marketplace add https://github.com/SukinShetty/Nemp-memory.git
/plugin install nemp

Manual install

cd ~/.claude/plugins/marketplaces
git clone https://github.com/SukinShetty/Nemp-memory.git nemp-memory
# restart Claude Code, then:
/plugin install nemp

Check it worked with /nemp:list - you should see your memories or "No memories saved yet".

Problems? Troubleshooting covers EPERM errors on Windows, commands not showing up, clone failures and clean reinstalls.

OpenClaw

Nemp also runs as an OpenClaw skill, using the same .nemp/memories.json.

git clone https://github.com/SukinShetty/Nemp-memory.git ~/.openclaw/workspace/skills/nemp-memory

On Windows the path is C:\Users\<you>\.openclaw\workspace\skills\nemp-memory. The repo includes the root SKILL.md OpenClaw needs. Restart OpenClaw and check that nemp-memory appears in your skills.

Save in Claude Code, recall in OpenClaw, and the other way round. OpenClaw uses the same AgentSkills standard, and Nemp only needs an agent that can read and write files.

Use cases

Onboarding. A new developer runs /nemp:init on day one and Claude already knows the stack, database, auth approach and structure.

Switching projects. /nemp:recall stack in ~/client-a says "Next.js, Stripe, PostgreSQL"; in ~/client-b it says "React, Supabase, Tailwind". Each project remembers itself.

Decision history. Save "REST, not GraphQL - team decision Jan 2026" today. Three months later /nemp:context api finds it, with no Slack archaeology.

Privacy

Everything stays on your machine. Nemp makes no network calls, sends no telemetry and needs no account. Your memories are files in your project and your home folder, readable in any editor.

Nemp Pro (coming soon)

Nemp Pro is in development and adds a memory intelligence layer on top of the free plugin, including cross-tool export and import for Codex CLI, Cursor and Windsurf. The Pro commands in this repo (/nemp:cortex, /nemp:foresight, /nemp:decay, /nemp:import) currently show a notice only. Follow the repo or nemp.dev for launch news.

Roadmap

• More stack detection beyond package.json (Python, Go, Rust)
• Svelte and Angular detection
• More reliable activity capture
• Cleaner skill instructions for AgentSkills-compatible agents

Ideas welcome in Issues.

Contributing

Framework detection, keyword mappings and suggestion rules are the easiest places to start. See CONTRIBUTING.md.

Support

• GitHub Issues for bugs and requests
• Email: [email protected]
• X: @sukin_s - share how you use it with #NempMemory

If Nemp saves you time, a star helps other developers find it.

License

MIT © 2026 Sukin Shetty. Free and open source.