How to Run a Second Brain on Free Local AI

By Austin Little

A "second brain" is just a place you trust to hold what your first brain should not have to: notes, decisions, lists, half-ideas, and the things you learned the hard way. You do not need a monthly subscription to build one, and you do not need to send your private notes to anyone's cloud to get AI help with them. A folder of plain text files, a free notes app, and a small model running on your own computer will do the job.

AI disclosure. This page was drafted with AI assistance and edited for Apiary. We don't invent quotes, stats, people, or events. Product details below were read off official Obsidian and Ollama pages on October 1, 2026. Software changes; re-check before you rely on any one setting.

What a second brain actually is

Strip away the productivity-influencer packaging and a second brain is three things:

1. A capture habit. You write things down when they show up, before they evaporate.
2. A home for notes. One place, not seven apps, where those notes live and can be found again.
3. A review habit. Every so often you look back, connect things, throw out what is stale, and turn notes into decisions or finished work.

That is it. A shoebox of index cards was a second brain. So was a farmer's weather notebook, a mechanic's binder of torque specs, and a beekeeper's hive log. The tools changed. The idea did not.

Where AI fits is mostly in step three. Review is the part people skip, because it is slow. A local model can help you summarize a month of notes, find the threads between them, draft a weekly review, or turn messy meeting notes into a clean list of next steps. It does that work in seconds, and if it runs on your own machine, the notes never leave home.

Why local, and why free

There are plenty of paid "AI second brain" services. Some are good. But they share a shape: your notes live on their servers, the AI runs on their servers, and access lasts as long as you keep paying and they keep operating.

A local setup flips that:

• Your notes are files on your disk. If the app disappears tomorrow, the files are still readable in any text editor.
• The AI runs on your hardware. There is no per-message meter and no monthly bill.
• Nothing leaves the machine unless you choose to sync or share it.
• You can swap parts. Different notes app, different model, same files.

The trade-offs are honest ones. You supply the computer. A small local model is less polished than a giant cloud model. And there is a little setup. This guide keeps the setup as small as possible.

The parts list

You need three things. All have a free path.

1. A folder of plain text notes

The foundation is not an app. It is a folder. Inside it, each note is a plain text file, usually written in Markdown, which is just text with a few simple marks like # for headings and - for list items.

Why plain text? Because it is the most durable format we have. Any computer made in the last few decades can open it. Any future tool will be able to read it. You are not betting your notes on one company.

2. A notes app that works on that folder

You can edit plain text files with any editor. A dedicated notes app makes linking, searching, and browsing easier.

Obsidian is a common choice for this kind of setup. Its official pricing page, as I read it on October 1, 2026, says it is "Free without limits" with "No sign-up required." Its optional paid add-ons are Sync and Publish, which you do not need for a local second brain. The page also says, in Obsidian's words, that "your data is stored locally on your device" and that "our apps do not collect telemetry data."

If you would rather not install anything new, a plain text editor and your operating system's file search still make a working second brain. The app is a convenience. The folder is the brain.

3. A local model runner

Ollama is a free tool that downloads and runs language models on macOS, Windows, and Linux. Its quickstart says that for local models, no API key is required.

As of October 1, 2026, the Ollama quickstart uses a small model called gemma4:e2b as its local example. It states the model download is "about 7.2 GB" and recommends "8 GB of available VRAM, or unified memory on a Mac."

Ollama also offers cloud models that need sign-in and an API key. For a private second brain, ignore those and choose Local.

The one limit you must understand: context

Before you set anything up, understand why a local model cannot just "read your whole vault."

Every language model has a context window: the amount of text it can consider at one time, measured in tokens. Ollama's context length documentation says its default depends on your graphics memory. On a machine with less than 24 GiB of VRAM, which includes most laptops, the default is 4k tokens. Machines with more graphics memory get larger defaults.

You can raise that number with a slider in the Ollama app's settings, or with an environment variable when you start the server. Ollama warns that a larger context "will increase the amount of memory required to run a model."

What this means in practice:

• A local model will not hold your entire notes folder in mind at once.
• You feed it one note, or a few short notes, per request.
• The structure of your notes matters more than the size of the model. Short, well-titled notes are easier to feed in.

This is not a flaw to work around. It is a design principle. A second brain built from small, focused notes is better for you and for the model. We cover context windows in depth in What Is a Context Window in Plain English elsewhere on Apiary.

Setting it up, step by step

Step 1: Make the folder

Create one folder. Call it something plain, like notes or brain. Inside it, make a few subfolders. Keep it shallow. Here is a structure that works for most people:

notes/
  inbox/        quick captures, unsorted
  projects/     things with a finish line
  areas/        ongoing responsibilities (house, health, garden, job)
  reference/    facts you look up again (recipes, specs, how-tos)
  journal/      dated daily or weekly notes
  archive/      done or stale, kept just in case

Do not overthink this. You can rename folders later. The only rule that matters is: everything new goes in inbox/ first.

Step 2: Open it in your notes app

In Obsidian, you "open a folder as a vault." In any other editor, just open the folder. Write a test note called inbox/first-note.md with a sentence in it. Confirm it shows up as a normal file in your file manager.

That last check matters. It proves your notes are yours, not locked in an app database.

Step 3: Install Ollama and pull one small model

Follow the official quickstart. Install the app. On Linux, the quickstart says to start the server with ollama serve if it is not already running. Then pull a small model and test it:

ollama run gemma4:e2b

Type a question. Type /bye to leave. If that works, you have a local AI.

Start with a small model. A model that fits comfortably in your memory will answer quickly. A model that does not fit will crawl. You can always try a bigger one later.

Step 4: Check your context setting

Run ollama ps while a model is loaded. Ollama's docs show this command prints a CONTEXT column with the allocated context length and a PROCESSOR column that tells you whether the model is on your GPU, CPU, or split. If your context is 4k and your notes are short, you are fine. If you want to feed in longer notes, raise the context in settings and watch whether the machine slows down.

Step 5: Connect notes and model the simple way

The simplest connection needs no plugins at all: copy and paste. Open a note, copy it, paste it into the Ollama chat, ask your question, copy the useful part of the answer back into a note.

That sounds primitive. It is also completely transparent. You see exactly what the model sees. Nothing is indexed, uploaded, or hidden. For many people, that is the whole system, and it is enough.

If you want tighter integration, Obsidian has community plugins that can talk to a local Ollama server.

Before installing any plugin, check three things: what endpoint it sends text to, whether it needs an account with a third party, and whether it can modify or delete files. A plugin that quietly sends notes to someone's cloud defeats the point of going local.

Daily use: capture, then let the model help sort

Capture fast, clean later

The capture rule is simple: get it down, do not format it. A phone note, a voice memo transcribed later, a line typed into inbox/. Do not stop to decide where it belongs.

Once a day or once a week, open the inbox and sort. This is where a local model can save time.

Prompt: turn a messy capture into a clean note

Paste the messy capture and ask:

Rewrite this as a short note. Give it a clear title, a one-sentence summary, and bullet points. Do not add facts that are not in my text. Mark anything unclear with [?].

That last instruction is important. Small models are eager to fill gaps. Telling the model to mark unclear spots rather than guess keeps your notes honest.

Prompt: suggest where it goes

Here is my folder list: projects, areas, reference, journal, archive. Which folder fits this note best, and why, in one sentence?

You make the final call. The model is a second opinion, not a filing clerk with authority.

Prompt: pull out action items

Paste meeting or phone-call notes:

List every action item in this note. For each, give who (if stated), what, and when (if stated). If no one is named, write "unassigned."

Then copy those into a project note or your to-do list.

Weekly review with a local model

The weekly review is where a second brain earns its name. It is also the step most people drop. A local model makes it lighter.

Because of the context limit, do it in passes rather than all at once.

Pass 1: summarize each day

For each journal note from the week, paste it and ask for a three-bullet summary. Save those summaries into a single journal/week-review.md note.

Pass 2: review the summaries

Now paste the summaries note, which is short, and ask:

Here are summaries of my week. What themes repeat? What did I say I would do that I have not mentioned again? Keep it under 150 words.

Pass 3: decide

Read what the model said. Then write, in your own words, three things you will do next week and one thing you will drop. That last part is yours. A model can point at patterns. It cannot decide what your life is for.

This "summarize small pieces, then review the summaries" pattern is the core trick for working with small context windows. It works for a week of journals, a stack of research notes, or a year of hive inspections.

Searching your second brain

People often expect the AI to be the search engine for their notes. For a free local setup, start with the search you already have.

• Your notes app's search finds exact words fast.
• Good titles do most of the work. "2026-09 roof leak, contractor quotes" beats "stuff."
• Links between notes let you follow a thread. In Obsidian, you link with double brackets, like [[roof leak]].
• Tags help for cross-cutting topics, like #taxes or #garden.

Where the model helps is after search. You find the five notes about the roof leak with normal search, paste them in one at a time or as short excerpts, and ask the model to summarize what you have learned and what is still open.

There is a more advanced approach where software automatically finds relevant notes and feeds them to the model, often called retrieval.

My advice: run the plain version for a month first. Most people discover that good titles plus normal search plus a model for summarizing covers what they need.

Privacy and safety habits

Local does not mean careless. A few habits keep your second brain safe.

Keep secrets out of notes, period

Do not store passwords, full card numbers, or Social Security numbers in plain text notes. Use a password manager for passwords. A notes folder is not a vault, and it may be backed up or synced to places you forget about.

Know where your folder syncs

If your notes folder sits inside a cloud-synced folder, your notes are going to that cloud, even if your AI is local. That might be fine. Just make it a choice. Obsidian's paid Sync service says it uses end-to-end encryption, per its pricing page, but you do not need it; a local-only folder with your own backups works.

Back up anyway

Local means you are responsible for backups. Copy the folder to an external drive on a schedule. Plain text files are small, so this is cheap and fast.

Be careful with tools that can act

A chat model that reads and writes text is low risk. A tool or plugin that can rename, move, or delete files across your vault is a different risk class. Test such tools on a copy of the folder first.

Treat model output as a draft

Small local models can misread a note, invent a detail, or summarize confidently and wrongly. That is why the prompts above say "do not add facts" and "mark unclear with [?]." Anything going into a decision about money, health, or law needs your own check against the source.

What hardware is enough

I am not going to print a universal RAM number, because the honest answer depends on which model you pick, and the right model changes every few months.

What I can say from the official docs:

• Ollama's quickstart example model recommends 8 GB of available VRAM or Mac unified memory, and says it can run slower using system RAM if you have less.
• Ollama's default context is 4k on machines with under 24 GiB of VRAM.

The practical test: install, pull the small example model, and ask it to summarize one of your notes. If the answer arrives in a reasonable time for you, your machine is enough. If it crawls, try a smaller model or keep requests shorter. If your computer cannot run any local model comfortably, a second brain still works with no AI at all. The notes are the brain. The model is a helper.

A sample week

Here is how this looks for one ordinary person: a part-time bookkeeper who also runs a small backyard garden and helps care for a parent.

Monday. Captures three things in inbox/: a client's question about a receipt, a reminder that tomato seedlings need hardening off, and a note from a call with the parent's pharmacy.

Wednesday. Spends ten minutes sorting. Pastes the pharmacy note into the local model and asks for action items. Gets back two items: call the doctor's office about a refill, and pick up on Friday. Moves the note to areas/parent-care/. The pharmacy details never leave the laptop.

Friday. Pastes the client note and asks the model to draft a short, polite reply asking for the missing receipt. Edits the draft. Sends it from regular email.

Sunday. Weekly review. Summarizes each journal note, then reviews the summaries. The model notes that "seedlings" appears three times with no follow-up. Writes a garden project note with a date.

No subscriptions. No sensitive notes on anyone else's server. About twenty minutes of AI-assisted work over the week, and a clearer head.

Common questions, answered plainly

Do I need to pay for a second brain app? No. A folder of plain text files is a complete second brain. Obsidian's official pricing page says its app is free without limits; its paid add-ons, Sync and Publish, are optional.

Can a local AI read all my notes at once? Usually not. Ollama's docs say its default context on machines with less than 24 GiB of VRAM is 4k tokens. Feed it one note or a few short ones at a time, or summarize in passes.

Is a local model private? A model running locally through Ollama processes your text on your own computer. Privacy then depends on the rest of your setup: where your notes folder syncs, which plugins you install, and who can use your computer.

What if my computer is too slow? Try a smaller model, keep requests short, or skip AI for now. The capture, organize, review habits work without any model at all.

Should I use cloud AI for my second brain instead? You can for non-sensitive notes. For private material like health, family, money, or client work, local is the calmer default.

What about my phone? Plain text notes can be synced to a phone using whatever sync you trust.

The bottom line

A second brain is a habit with a folder attached. Capture quickly. Keep notes small and well titled. Review on a schedule. A free local model, run through a tool like Ollama, makes the review and sorting faster without sending your private life to anyone's server.

Start with plain text, one small model, and copy-and-paste. Add plugins and automation only when you can name the job the simple version is failing at. Your notes should outlast every app you ever use to read them. Plain text on your own disk is how you make sure they do.

References

• Obsidian — Pricing — https://obsidian.md/pricing (fetched 2026-10-01)
• Ollama — Quickstart — https://docs.ollama.com/quickstart (fetched 2026-10-01)
• Ollama — Context length — https://docs.ollama.com/context-length (fetched 2026-10-01)