FAQ

FAQ

Where does this run, and does my data leave my machine?

It runs locally, on your own hardware. The vault is a folder of Markdown files on your disk, the scripts are Python, the schedule is launchd or systemd on the same machine, and the search index is built in place. There is no hosted service, no account and no sign-up. The only thing that ever leaves the machine is a prompt to whichever model you configured; point BRAINLESS_LLM_PROVIDER at Ollama or LM Studio and even that stays at home. See Configuration and Ownership and Privacy.

What problem does it actually solve?

Two. The first is that notes end up scattered across voice memos, screenshots, e-mail and chat windows. Written down somewhere and then effectively gone, because "I wrote that down" and "I can find it again" are not the same sentence. The second is worse: decisions made on a phone between two meetings and never looked at again, which is how you get to be wrong the same way twice without noticing. A pile of notes does not fix either one.

So what is the solution?

One loop, and each part is there because the one before it is useless alone. It works in four lines:

1. Gather information. Everything comes in through one door, nothing judged on the way in. The machine reads it, summarises it and makes the links for you.
2. Think deeper and clearer. Ask what you already know before anything hard, then let six personas argue with the thesis in two rounds.
3. Decide. Write the decision as a bet, with a prediction, a confidence and a review date, and settle the bet when the date comes.
4. Bonus: get shit done. Promises become tasks, the morning brings at most three things, and you answer reminders in a thread.

The README tells it in one page; How It Works has the mechanics.

How do I use it day to day?

• All day, with no effort. Send voice notes, photos and links to the capture bot on your phone. That is the whole capture ritual. No folders, no tags, no inbox to process later.
• Before anything hard. Run brainless search or a slash command to find what you have already written on the subject, including the parts you had forgotten writing. Before a real decision, run brainless dialectic "<thesis>" and read what comes back.
• Morning. The night is already compiled. The Today Queue offers at most three items: a decision due for grading, a commitment, one piece of evidence.
• Overnight, without you. The worker digests the day, rebuilds the compiled layer and runs the personas at 12:30 and 21:20.

Does it help me get things done?

Yes, that is the fourth line. Promises you make in a meeting or a voice note are picked up as tasks, checked against the list so the same job is not written down three times in three wordings, and synced with Google Tasks both ways. Each morning brings at most three things, each with a link to where it came from. You answer in a thread on your phone: "apply" writes the note, "defer to Friday" moves it, or you drop it with a reason. See Today Queue and Buzz Interactions.

Do I need three devices?

No. One laptop is the baseline and runs the whole loop with none of the addons. The phone is for frictionless capture; the always-on worker is where you go when the Mac being closed starts to cost you. See Reference Deployment.

Do I need a Claude subscription?

The default path assumes the claude CLI signed in on a subscription, so there is no API key to manage. Any OpenAI-compatible endpoint works instead (OpenAI, Together, Grok, Ollama, LM Studio), though it has been tested less. See Configuration. If your plan is Max or Team, it includes a monthly API credit: the night shift can run on it through the anthropic provider, with a ledger and a budget cap, while your own sessions stay on the subscription. See The AI Shed.

Will it write into my notes?

Not without you. The machine proposes and you apply: a preview waits in its thread until you answer "apply". Capture workers own only new files in Thinking/Daily/ and Inbox/Links/. The text-processing LLM never gets general file tools. See Ownership and Privacy.

What stops my private folders being compiled?

Identity papers, health and official documents are ignored by default. Add any folder name to private_segments in _Agent-Context/PROFILE.md and it is never read. The public repo is produced by an export script with a whitelist and a leak scan.

Can I change the personas?

Yes, and you should. Each persona's prompt is a Markdown file in .agents/buzz/personas/. Edit one, add a seventh, delete one. The moderator does not care how many there are. See Personas and Dialectic.

Can it work in another language?

output_lang in PROFILE.md sets the language of every LLM output; any language the model can write works. Fixed labels and bot replies ship in English and Turkish; another language is a new locale directory. See Localization.

How does it make me think clearer?

By what it makes you write down before you may move on, and in what order: criteria before options, a base rate before a confidence, six readings taken blind, one dated action at the end, a red line when all six agree. The full list with the file each one lives in is on Thinking Clearer.

Does it make decisions for me?

No. It never picks. It captures, compiles, argues and grades. Choosing is yours; a system that chose for you would be resulting in advance.

Where does the output go?

Every analysis, graded or not, is filed into .wiki/digests/queries/, so tomorrow's question can build on today's answer. The human notes remain the source of truth; .wiki/ is regenerable and never hand-edited.

Who is it for?

People who make decisions for a living and suspect their own reasoning. Founders, operators, product people, anyone whose calendar is a sequence of judgement calls. It helps if you already write things down. It helps more if you are willing to be told you are wrong by a script at 21:20.

Why is it called brainless?

It is a joke that stuck. "Let us use our own brain first, then we can talk about a second brain." So not a second brain, zero brain. It turned out to carry the design rule: the system keeps the memory and the connections, the thinking stays yours.

Where do the ideas come from?

Books, papers and tools, each with what was taken and where it shows up in the code: References.

How do I run the tests?

They run offline, with no credentials and no model calls:

python3 -B -m unittest discover -s tools/tests -v

How do I contribute?

Issues and pull requests are open on the repo. If a persona gave you a bad argument, paste the thread. If the installer broke, paste the output. Both are more useful than a feature request.

Honest limits

• The default path assumes a Claude subscription. The OpenAI-compatible path is tested less, and reading handwriting and whiteboard photos only works with Claude for now. Text, voice and documents are fine.
• Fixed labels and bot replies ship in English and Turkish only.
• The six personas are one person's shelf. Yours may differ, and should.
• It was built for one person, then made installable in a day. Expect rough edges, and say where they are.