Build an AI Second Brain: A Zettelkasten That Maintains Itself

Everyone who has tried to keep a second brain knows how it ends. The method is fine. What breaks is the upkeep: reading every source, writing it up, deciding where it goes, noticing the old note it contradicts, merging the duplicates. That's a second job, so after a few weeks the inbox folder is full, the notes stop growing, and the system goes stale.

This guide keeps what makes a Zettelkasten valuable: small notes, one idea each, organized by topic so your thinking builds up over time. The upkeep goes to two workflows in Damon. One runs every evening and captures what happened that day from your email, meetings and chats. The other runs every week and keeps the whole thing in shape. The notes are plain Markdown in the Damon drive, and nothing leaves your workspace.

What you get

A Brain/ folder that stays current without you tending it, and an evening message like this:

2 meetings, 3 flagged emails and 4 chat threads captured. New notes: Annual prepay cuts churn more than monthly discounts (from the Northwind renewal call), Onboarding stalls when the first integration fails silently. Updated: Enterprise buyers ask for SSO before pricing, now with a third source. The Onboarding topic has enough notes for its own map, so it has one.

On Sunday a second message says what the weekly upkeep did: two duplicates merged, one note that had grown to cover two ideas split, a map summary rewritten because the notes under it changed.

Ask Damon "what does @Brain say about annual pricing?" on any day and the answer comes from your own notes, with the sources cited.

The structure

A Zettelkasten has a few kinds of notes, and each gets a folder:

Brain/ README.md        your conventions; the agent reads them before every run Inbox/           quick captures you jot down yourself Sources/         one note per email, meeting or chat, summarized Notes/           permanent notes: one idea each, in plain language Maps/            one note per topic: current thinking and the notes behind it

Sources are the raw material, named YYYY-MM-DD-<source>-<title>.md, e.g. 2026-09-29-meeting-northwind-renewal.md. Each has a short summary, the key points, and where it came from (sender and subject, meeting title, or chat).

Notes are the core of the system. Each holds one idea, titled as a claim: 202609291830-annual-prepay-cuts-churn.md, titled Annual prepay cuts churn more than monthly discounts. "Pricing" is a topic; that title is an idea, and ideas are what you want to find again. A note lists the sources behind it, so every claim can be traced back:

# Annual prepay cuts churn more than monthly discounts

Customers who prepay for a year renew at a higher rate than customers given a monthly discount of the same value; the commitment matters more than the price.

Sources:
- 2026-09-29 meeting: Northwind renewal
- 2026-09-12 email: Q3 churn review

Maps are how you find your way in. Each covers a topic, like pricing, onboarding or hiring, with a paragraph on where your thinking stands and the notes that support it. Maps are where the second brain changes over time: as notes pile up, the summary paragraph gets rewritten to match.

Step 1: Connect Gmail and Fireflies

In Integrations, connect Gmail and Fireflies. The workflows only read from them: searching and reading mail, listing and reading transcripts.

In Gmail, create a label called Brain. That's how you capture: when an email is worth keeping, apply the label. The agent reads labeled mail, not your whole inbox.

Step 2: Write the conventions

You don't need a special agent for this. Damon, the default agent, does the work, and the rules live in the folder itself. In chat, ask Damon:

Create a Brain folder with Inbox, Sources, Notes and Maps inside. Write Brain/README.md with the conventions: how source notes are named; permanent notes hold one idea each, titled as a full-sentence claim in plain language, with a timestamp ID and a list of sources; update an existing note rather than creating a near-duplicate; quote sources exactly and never invent; a map for any topic with three or more notes.

The README is your spec. Change your mind about the system, edit the README, and every run after that follows the new version.

Step 3: Capture one day by hand

Using the conventions in Brain/README.md, read my emails labeled Brain from the last 24 hours, my Fireflies meetings from today, and anything in Brain/Inbox. Look back over what we discussed in chat today. Write a source note for each, then create or update permanent notes and maps. Tell me what you changed.

Writing to the drive asks for approval by default, so this first run checks with you before each write. Read what it produces. Most first-run fixes are about granularity ("that's two ideas, split it") and naming, and they belong in the README.

Step 4: Capture every evening

1. Trigger: every day at 18:00.
2. Agent step: reads Brain/README.md, then the day's sources:

• emails labeled Brain from the last 24 hours,
• Fireflies transcripts from the last 24 hours,
• anything in Brain/Inbox/,
• what came up in your chats with Damon that day.

It skips anything that already has a source note, writes the new source notes, then updates the permanent notes and maps. A new idea gets a new note, and a new source for an existing idea is added to that note.

3. Message step: tells you what was captured, which notes are new or changed, and any new maps.

Because every source gets a note, a source that's already filed is skipped next time, so a re-run never duplicates anything.

For this workflow, auto-approve writing to your drive so the evening run doesn't wait on you. It still only reads Gmail and Fireflies, and deleting files always asks.

Step 5: Upkeep every week

Capture keeps the brain full. What keeps it useful is the upkeep a person would do on a slow Sunday and never actually does. Add a second workflow:

1. Trigger: every Sunday at 10:00.
2. Agent step: reviews Brain/:

• merges notes that say the same thing,
• splits any note that has grown to hold two ideas,
• retitles notes whose claim no longer matches their content,
• flags notes contradicted by newer sources,
• rewrites each map's summary to match its current notes,
• creates maps for topics that have grown enough,
• moves notes whose newest source is over six months old, and that no map uses, to Brain/Archive/.

3. Message step: what changed, and anything it wants your judgment on, such as two notes that disagree.

Start with the upkeep in report-only mode ("list what you would change; change nothing") for a week or two. Once its suggestions are the ones you'd make, let it make the changes.

Using it

• Ask it. "What does @Brain say about enterprise onboarding?" Mentioning the folder points Damon at your notes. It searches their contents by meaning, not just file names, and answers with sources.
• Jot things down. A half-formed thought goes into Brain/Inbox/ as a line or two. That evening it becomes a note, or part of one.
• Read the maps. They're the table of contents you never had to maintain. When a map's summary surprises you, your thinking has shifted, and the upkeep caught it.

Variants

• Slack as a source. Connect Slack and add "messages I've reacted to with :brain: in the last 24 hours" to the capture step.
• Team brain. Point both workflows at a shared folder and the team's meetings. The maps become the team's shared picture of each topic. For the product-feedback version, see feedback synthesis for product managers.
• Meetings only. If meetings are most of your input, start with the meeting follow-ups recipe, which already files every meeting to the drive.

Common questions

Does it read my whole inbox?

It reads the emails you label Brain. It can search any mail the connected account can, so if you want a hard boundary, connect a separate mailbox.

Will it send or change anything outside the drive?

No. Both workflows only read Gmail and Fireflies and only write to Brain/.

What if it files something wrong?

Every note lists its sources, so you can check any claim against the original. Tell the agent what was wrong and put the rule in Brain/README.md; the next run follows it.

Isn't the point of a Zettelkasten that you write the notes yourself?

Writing ideas in your own words is how the method builds understanding, and nothing here stops you. Edit any note, or write in Inbox/ and let the agent handle only the filing. What the agent takes off your plate is the upkeep, which is the part that makes most people quit.