A model is four things. Libro is the other four.
Weights, a tokenizer, an inference loop, some marketing — that's a model. Libro is the rest: memory you can walk, an index you own, a harness that routes each question, an orchestration layer across models.
You bring your Claude. Libro brings the system around it.
Data → Memory → Decision → Output
Four layers. All on your machine, in files you can read.
Yours, staged
Notes, docs, dictation, transcripts — into one brain folder you own.
Two tiers
Obsidian vault for thinking, local vector store for retrieval. Neither hosted.
You set the ceiling
Routes which model answers which question — under a budget you control.
Lands where work lives
Markdown, slides, briefs — back into your vault or terminal.
Four things the big platforms won't let you own
You own the index.
Chunking, metadata, embedding model, retrieval weights — all yours, all inspectable. Not a vendor black-box.
You own the harness.
Your Brain folder is the harness. What's captured, when it's retrieved, how it's weighted — your rules.
You own the orchestration.
Route tasks across model tiers by cost and match — cloud or local. Budget visibility from hour one.
You bring the intelligence.
Libro ships no models — it ships the system yours run inside. Your Claude, your local Ollama, Libro around them.
Free. And it stays on your machine.
Libro is free. Not freemium. Not trial. Your data never hits our servers — there are no servers. Files you can read, paths you can grep. Uninstall tomorrow and your data stays put.
We charge for BlackBox — the on-premises stack — only when "your stuff never leaves your building" is a hard requirement. Until then, Libro on your laptop is what you get.
Trustworthy by default, not by promise.
Reversibility
Every change undoable. Libro supersedes, never overwrites. Roll back without a vendor ticket.
Observability
No silent subprocess. Every state change logs before it fires. You see what Libro did and when.
Reproducibility
Same inputs, same outputs. Trust tags on every claim from outside your brain.
Install Libro. Keep your stuff.
Open-source under Apache 2.0. Read the script before you run it — that's the point.
Build-in-public updates
Building an AI-native company in the open. New skills, what broke, what shipped. No spam.
Straight answers.
What is Libro?
Libro is a free, open-source AI operations framework for Claude Code. It gives you operator-owned memory, an index you control, a harness that routes each task, and orchestration across models — the system around the model, not the model itself. Licensed Apache 2.0.
Is Libro free?
Yes. Libro is free and open-source under Apache 2.0 — not freemium, not a trial. You bring your own Claude subscription; Libro is the framework you run it inside.
How is Libro different from an AI model?
A model is weights, a tokenizer, and an inference loop. Libro is the other half: memory you can read, an index you own, a harness that routes each question, and orchestration across model tiers. Libro ships no models — it ships the system yours run inside.
Does Libro send my data anywhere?
No. Libro runs on your machine — there are no Libro servers. Your memory is local files you can read and grep; your vector index stays on your hardware. Uninstall tomorrow and your data stays put.
What do I need to run Libro?
macOS or Linux, a Claude Code subscription (bring your own key), and optionally a local Ollama install for offline model routing. Install from GitHub under Apache 2.0.
Who builds Libro?
Libro is built by Cerebro Cyber Solutions, a Service-Disabled Veteran-Owned Small Business (SDVOSB), on one thesis: the operator should own the memory.
Learn the ideas behind Libro.
What is a Claude Code framework?
How memory, skills, and model routing turn bare Claude Code into an operator-owned system — and where Libro fits.
AI that remembers you
Why most AI forgets between sessions, and how local AI memory keeps your context in files you own.
Own your AI vs. rent it
What operator-owned AI means in plain English, and why owning your memory, index, and routing beats renting a black box.