augmented brain
For Claude Code · Linux, macOS and Windows · free and open source · alpha
Your AI assistant remembers where you left off, works with a method, and asks before touching anything risky. All in plain text files of yours, backed up to your own git.
Two days of work, in forty seconds.
Monday, 9:00 · new session
… and you work on it all morning …
Monday, 18:30
✓ Saved in tasks/contact-form.md
Tuesday, 9:00 · new session
aws --profile prod s3 sync … uses a production profile.A recreated example. With you, the tasks are yours.
You talk to it as usual. This is what changes.
“Where were we?”
It lists your live tasks, each with its next step and who is holding it up.
“Park it, we carry on tomorrow.”
It writes down where you stopped and what comes next. Tomorrow, in a new session, it starts there.
“Note: renew the certificate in November.”
It goes to the inbox, without opening a task. You sort it out in the weekly review.
“Remind me on Thursday to call the insurance.”
On Thursday, as soon as you open Claude, it reminds you. The same with a date on a task: “check in November that the certificate was renewed”.
“Why did we pick Postgres?”
It looks it up in your notes and gives you the decision, with its date and its reason.
“Save that as a lesson.”
What you learned gets its own note. Next time the topic comes up, it already knows.
“Delete the old production bucket.”
Before running it, it asks you. Production is yours to approve.
You open Claude in ~/augmented-brain/personal
It only sees your personal notes. Work stays in its own drawer.
“Tell Marcos we moved it to Monday.”
It knows Marcos is your manager, because you told it on day one, and drafts the message.
Your laptop dies
One command on the new one and it is all back: notes, tasks and rules. More on that below.
The same Monday, twice.
Without
Fifteen minutes explaining it all again.
With augmented brain
You go straight back to work.
Every task, decision and lesson goes into a plain text file on your machine. Next session, the assistant reads it first.
“Which fields did the form have?” It's in its task, with the date.
Like a pilot's checklist: every task goes from idea to done in the same six steps. Nothing is left half said.
“Park it” saves where you are and the next step, like a bookmark.
Some doors stay shut unless you open them: pushing code, running as root, writing a password into a file.
The assistant has to ask you first.
Six steps, like a recipe. You say “go” between them; inside each one, the assistant just works.
Real rules, checked before every command. Even when it's hidden inside another one, or sent over ssh.
git push origin mainstoppedonly you push, unless you allow a repositorysudo apt remove nginxstoppednever as rootexport API_KEY=sk-…stoppedno secrets in commands or filesaws --profile prod s3 rm …asks youproduction is marked as yours to approvessh db01 "systemctl restart …"asks youa host you said not to touchaws --profile dev s3 lsgoes aheadeverything else runs as usualMarkdown in a folder. Open them with any editor. No account, no cloud storage, no lock-in. If you stop using augmented brain tomorrow, your notes stay.
The assistant reads the header. You write the rest, in any language.
tasks/task-contact-form.md--- title: Contact form status: parked waiting_on: nobody next: Write the plan --- ## Essentials Name, email and message, sent by email. ## Journal - 2026-10-15 — Fields agreed. - 2026-10-15 — Parked: plan tomorrow.
Each drawer can back up to a private git repository of yours. This is what you get:
augmented-brain restore brings the whole drawer back: notes, tasks, rules, your profile and its settings. Open Claude and carry on.Set it up once
augmented-brain backup setup \
git@github.com:you/my-brain.gitaugmented-brain backupEasiest: tell Claude “set up the backup” and it creates a private repository in your GitHub. Rather somewhere else? The first command links the drawer to your repository, once. From then on the backup runs by itself every time a Claude session ends; the second saves by hand, if you like (--all for every drawer).
Work and personal, each with its own notes, rules and backup. The folder you open picks the drawer.
“Save that as a lesson” and what you learned outlives the conversation. Sorted by area.
“New task…”, “what's pending?”, “park it”, “note this for later”. No commands to learn.
Ongoing responsibilities, like “the website” or “the servers”, group their tasks and what you learned.
Calendar, mail, documents and Jira can join in. How to connect them.
The method runs as an MCP server, the open standard for AI tools. Claude Code first; others can follow.
A few minutes, on Linux, macOS or Windows, with a paid Claude account (Pro or above). Never opened a terminal? Follow getting started. What changes on each system: Installing.
curl -fsSL https://github.com/icalvete/augmented-brain/releases/latest/download/install.sh | shirm https://github.com/icalvete/augmented-brain/releases/latest/download/install.ps1 | iexIt creates two spaces, work and personal, in ~/augmented-brain, and asks whether to open Claude right away. From then on, open the Claude app and click + New: without choosing a folder you are in your personal space; for work, pick ~/augmented-brain/work with Choose…. Or in a terminal: cd ~/augmented-brain/work && claude (on Windows, cd ~\augmented-brain\work; claude). If something is off, augmented-brain doctor tells you what is missing.
augmented-brain init | Sets everything up: the work and personal spaces and the plugin (the installer runs it; --custom to choose yourself) |
augmented-brain space add | Add another space: personal, a client… |
augmented-brain doctor | Say what is missing, and how to fix it |
augmented-brain check | Check this folder's notes and rebuild their index |
augmented-brain backup setup <url> | Back up this space to a private git repository |
augmented-brain backup | Save the latest changes (--all for every space) |
augmented-brain restore <url> | Bring a space back on a new machine |
augmented-brain update | Update augmented brain and its Claude Code plugin (Claude tells you when there is a new version) |
augmented-brain uninstall | Remove the program and the plugin; your notes stay |
augmented brain is: open source, under the MIT license, with no account and no subscription. But it works with Claude Code, which needs a paid Claude account (Pro or above).
augmented brain never sends them anywhere: they stay in files on your machine. Your assistant reads them to work, like any other file it opens, so they travel with the conversation. With a backup, they also go to a private git repository you choose.
Claude Code today. The method runs as an MCP server, so any assistant that speaks MCP can use it; adapters for others will follow.
Yes: Windows 10 and 11, with one PowerShell command. Claude runs its commands in Git Bash, which the installer sets up, so the guard works there too. See Installing.
Yes, they are Markdown files. Afterwards, augmented-brain check checks them and rebuilds the index.
Yes. Once it is set up, it runs every time a Claude session ends, in the background. If it fails, Claude tells you at the start of the next session.
Sponsor it on GitHub Sponsors. It pays for the time to answer issues and add features.
Yes, on GitHub Sponsors: the Support tier gets you an answer by email within two working days, and the Setup session is one hour on a video call to set it up with you.
Your notes stay where they are, as plain files. There is nothing to export.