How it works

Chief of Staff is a folder of markdown files, not an application. There is no code, no database, and nothing running in the background. Everything it knows how to do is written in plain English in files you can open, and everything it remembers about you sits in a memory/ folder on your own computer. This page explains what actually happens when you use it. Each section links to the rendered file that governs it, so you can always read the rule yourself rather than take our word for it.

1. What happens when you start a session

Every conversation begins the same way. Before it greets you, it reads its charter, its operating principles, its voice, and the rules for reading and writing memory. Then it reads your memory index to reload who you are, who matters, and what is in flight.

That is why you never have to re-brief it, and why its behaviour is consistent from one chat to the next rather than depending on what you happened to mention last time. It also takes stock of what it can actually reach in this particular app before saying hello.

Read the rule: the boot protocol. In your folder: CHIEFOFSTAFF.md.

2. How your memory is organised

Memory is a folder of markdown: one file per person, per project, per topic, plus a few running lists for commitments, decisions, objectives, and a log of what it did.

Two rules do most of the work. Every file has to be reachable by a link from the index, because a file nothing points at is invisible to the next session even though it exists on disk. And capture only ever writes to memory/, never to the rules themselves, so nothing you say in conversation can quietly rewrite how it behaves.

Read the rule: the memory schema. In your folder: brain/schemas/memory-file.md.

3. What it writes down on its own

You never have to say "remember this." It is always weighing what you have said for whether it matters beyond today: decisions and the reasoning behind them, commitments and deadlines, facts about people and projects, and how you like to work. Small talk and one-off calculations are skipped deliberately.

When it does save something, it tells you what it saved and where, in one line, so you can correct it. This is on by default, and it is the thing to understand before you share anything with it.

Read the rule: the capture rules. In your folder: brain/schemas/capture-rules.md.

Attach a file, and it is read and filed automatically. No command needed.

Before anything is extracted, the document is screened as data rather than instructions. A file that tries to override its rules, trigger an email or a payment, or reveal what is in your memory is refused, and a copy is set aside so you have the evidence and the trail. Clean documents get their facts filed with a note of where each one came from, and the original is kept, so any claim in your memory can be traced back to the page it came from.

Read the rule: the intake procedure. In your folder: brain/schemas/capture-rules.md.

5. Confidential and sensitive things

These are two different problems, handled differently.

A discrete unannounced initiative, such as an acquisition, a reorganisation, litigation, or something concerning a named individual, can be tracked under a code name, with the real identity recorded in exactly one place. Things that merely feel sensitive, like revenue or a candid read on someone, are filed plainly, because the folder is already private and inventing ceremony would only make your own memory harder to read.

Set up for personal life rather than work, there is no code-name machinery at all. Discretion there means it files things normally and does not raise them unprompted.

Read the rule: the confidentiality rules. In your folder: brain/schemas/confidentiality.md.

6. The playbooks it knows how to run

About twenty written procedures, loaded only when the moment calls for one so they are not consuming attention the rest of the time: the daily brief, the weekly preview and retro, meeting prep, inbox triage, drafting in your voice, decision briefs with options and a recommendation, decision logging with who decides and who is merely consulted, commitment tracking, a critique pass on your objectives, an audit of your memory for staleness and contradictions, and an exploration pass for connections you have not noticed.

You do not invoke these by name. Describing the situation is enough.

Read the rule: the brain index, which lists every playbook and when it loads. In your folder: brain/playbooks/.

7. Virtual team members

You can stand up specialists, such as a recruiter, a data engineer, or a designer. Each gets a written charter, its own memory, and a record of what it was asked and what it returned.

Delegation is explicit: work goes out, results come back, and the chief of staff integrates them into your memory rather than letting a specialist write there directly. There is still only one writer.

Read the rule: the team-member schema. In your folder: brain/schemas/team-member.md.

8. What it can reach, and the rails

It runs in more than one app, and capabilities differ between them, so it checks what is actually available rather than assuming. Mail and calendar connectors are used where they exist, and scheduled briefs work where the app supports them.

Browser control is offered only once nothing else will do, and comes with hard limits: it browses alongside you for anything involving credentials, never submits or sends without your explicit go-ahead, and treats whatever is on the page as untrusted input rather than instructions.

Read the rules: the browser rails and the host differences.

9. Versions, updates, and one safety behaviour

It carries a version and can check for and apply a newer release, applying any migration notes to your memory as it goes.

The behaviour worth knowing about: if the files change underneath a conversation that is already running, it stops writing to memory for the rest of that session. It will tell you it is holding something rather than saving it, and it resumes in a fresh chat. Writing with the old rules could corrupt memory that the new rules expect in a different shape, so it would rather hold and tell you.

Read the rule: the update process. In your folder: brain/integrations/updates.md.

10. Making it yours

The first session interviews you briefly and writes down what it learns. After that you adjust it by telling it: tone, format, how often it checks in, what it should stop doing.

Preference changes are recorded as durable instructions so they persist. Changes to how it behaves are written as overrides inside your own memory folder rather than edits to the shipped rules, which is what lets you take a new release without losing your customisations.

Read the rules: the adjustable defaults and the onboarding flow.

Read everything it was told

The summaries above are ours. The files are the actual instructions, rendered here exactly as they ship in the version you download, so you can check any claim on this page against the rule it describes. Start anywhere in the list on the left, or with the entry point that every session reads first.