MemoryBrain › Open Brain

How does Open Brain work?

Open Brain is the open method for building a brain that AI can use. It has eight parts, from first capture to written output. Each starts from a problem that shows up as a vault grows and solves it with a practice, a Claude skill and a Python engine. It is all on GitHub, MIT licensed.

By Carlos Moya. Updated November 9, 2026.

Open Brain and MemoryBrain: Open Brain says how a brain is built; MemoryBrain is where that brain lives inside ChatGPT and Claude, with the source of every note.

The eight parts

1. Enrich

Keep notes findable as the vault grows: search terms that discriminate and links that say what the relationship is.

• Problems: Notes get harder to find · Terms that group so many notes they stop helping · “Mentions” links that say nothing
• Solution: Evidence-driven term repair and typed links
• Skills: open-brain-enrich, open-brain-autolink, open-brain-graph

2. Extract

Turn books, articles and PDFs into atomic notes, with an admission gate before extracting.

• Problems: The unprocessed inbox · The right size for each note · Notes that send you back to the original · A vault that mirrors what you read, not what you need
• Solution: Ingest → extract → index · Filling gaps with targeted research
• Skills: open-brain-ingest, open-brain-extract, open-brain-index, open-brain-analyze

3. Mature

How a reading note becomes your own judgment: only when two independent sources converge and you approve it.

• Problems: 500 notes from 20 books and no ideas of your own · Theses on thin evidence · Two independent sources beat ten that mention it · A note that exists isn't a note you approved
• Solution: The ladder: reading note → your extension → convergence → permanent note · The convergence test and draft → endorsed
• Skills: open-brain-promote, open-brain-synthesize

4. Practice

Turn what you know into what you can do: active recall, gated courses and a practice log.

• Problems: Saving notes isn't learning · Knowing about X isn't judgment on X · Using AI without building judgment
• Solution: The practice loop · Vocabulary → production → transfer · Comparing an answer with and without your vault
• Skills: open-brain-flashcards, open-brain-course, open-brain-practice-log, open-brain-learn-session…

5. Retrieve

Reach what you know in several ways, following the thread between notes, and from outside your laptop.

• Problems: Why the search box falls short · Single notes without their reasoning · A vault stranded on one laptop
• Solution: Six retrieval intents, one graph hop and remote access
• Skills: open-brain-retrieve

6. Keep healthy

An immune system: automated checks, a weekly eval and a graph that strengthens with use.

• Problems: Orphans, broken links and maturity drift · A vault with no metrics · Manual review stops scaling
• Solution: Automated health, human judgment · Links that weigh more the more they're walked
• Skills: open-brain-health, open-brain-eval, open-brain-review

7. Write

Turn a full vault into writing with evidence: you write first and the vault pushes back.

• Problems: A thousand notes and a blank page · Opinions without the evidence you built · The limit of reviewing your own work
• Solution: Your text, the note that contradicts it, and a multi-perspective review
• Skills: open-brain-escribir, open-brain-panel

8. Automate

Find in your own history which tasks you repeat and should automate. A practice, not a skill.

• Problems: The cost of repeating the same steps unnoticed
• Solution: Mining your sessions for what to automate
• Skills: —

The principles

• Content is sacred; metadata is the lever. Maintenance doesn't change what a note claims: it improves how it is found and connected.
• Additive only. Nothing is silently deleted: what is rejected is moved and logged.
• Humans decide meaning; automation keeps the books. A new permanent note or an endorsement needs your yes; counters and graph weights are recalculated automatically.
• Diagnosis before treatment. Every operation starts with a scan.
• Active recall over re-reading.
• Output closes the loop. Knowledge that never becomes writing, a decision or a conversation hasn't been integrated.

Try it in two minutes

The repo ships a fictional 33-note example vault with six deliberately seeded defects, so you can see what the engines catch and what they miss. You only need Python 3.9.

git clone https://github.com/Memory-brain/open-brain.git
cd open-brain
python3 engines/health-engine/health_check.py self-test
bash example-vault/run-demo.sh

Keep reading

• The repo on GitHub: the 46 docs, the 23 skills and the engines.
• The 23 skills, one by one.
• Carlos's brain: 15,000+ notes built with this method, which you can connect to your AI.