Your AI tools forget you every conversation. Munkymind is an open-source personal context library: point it at your notes, and Claude, ChatGPT and other tools can draw on them whenever they need to.
One command to set up. Needs Docker and one OpenAI API key. Setup costs a few cents.
Markdown, text, PDFs, an Obsidian vault as is, and GitHub repos you choose
On your machine: a searchable library of you, with sources and freshness
Claude Desktop, claude.ai, ChatGPT and other MCP clients
Your library stays on your machine. Note text goes only to the AI provider you choose, under your own key, or nowhere at all with Ollama.
Self-host with Docker. Your library is stored on your computer, and only reachable from it unless you choose to connect a remote tool.
Note text is sent only to the AI provider you choose (OpenAI or Anthropic), under your own key, to index it and write answers. Want nothing to leave your machine? Use Ollama, with no key at all.
Give an agent or tool a read-only key, make it expire in 30 days, rotate it with one command. Expired keys are refused everywhere, including by connected AI tools, and a rotated key stops working at once.
Runs as an unprivileged user with every Linux capability dropped, and every release is scanned for known vulnerabilities before it ships.
Apache 2.0 licensed. Read the code, change it, add your own connectors.
You need Docker, git and one OpenAI API key. Indexing 50 notes takes a minute or two and costs a few cents of OpenAI usage; each question costs a fraction of a cent. Full checklist: Before you start.
git clone https://github.com/munkymind/munkymind.git
cd munkymind
./setup.sh
It asks for your key (and checks it works), a username and your notes folder, then starts everything, indexes your notes and runs a test question. Ports already in use? It picks free ones. Windows: run it from WSL or Git Bash. No notes handy? Answer ./examples/sample-notes to try a small made-up library.
mm_sk_… key it prints (it is shown once), then:
docker compose exec api munkymind query --user yourname "What am I working on?"http://localhost:8000/ui) and paste your key to browse what it learned.A local viewer at localhost:8000/ui. Every topic is a brain (Body Brain, Business Brain, Big Picture Brain), and your AI tools are their senses. Browse what each brain knows, and see where it came from.
Type a question and see exactly which few snippets your AI would be sent, and how small that is next to your whole library. No AI is called.
Brains get hungry. Each one shows a mood, from 😋 fed to 😴 starving, so you know which notes are out of date before your AI relies on them. And they feed themselves: edit, add or delete a note and it's picked up within minutes, re-indexing only what changed.
“Given my goals and this week, what are my top three priorities?” One answer that draws on work, strategy and calendar, with sources.
“Plan my runs this week around my calendar.” Your plan, your rules and your week, without re-explaining them.
“Prep me for my 1:1 with Priya: open actions and what she raised last time.”
Connect GitHub, then ask Cursor or Claude Code: “Which of my repos already solve auth?”
What can I do with it?: a 15-minute tour, use cases and example prompts.
Munkymind is in early testing (latest release). Install it, try the tester missions with your real notes, and tell us how long it took and where you got stuck.
Get an email when there's something new: releases and major features only. Optional, and unsubscribe any time. You don't need it to use Munkymind.
We only use your address for Munkymind updates. Unsubscribe any time.