Atlas Cloud is a full-modal AI inference platform that gives developers a single AI API to access video generation, image generation, and LLM APIs. Instead of managing multiple vendor integrations, you connect once and get unified access to 300+ curated models across all modalities. Check out Atlas Cloud's new coding plan promotion for more budget-friendly API access: https://www.atlascloud.ai/console/coding-plan
For an example of what Second Brain can do, visit https://second-brain.art! It's an interactive art exhibition.
Second Brain performs higher than OpenClaw and Hermes on harness-bench, which measures "model-harness configuration effects across 106 sandboxed offline agent tasks". Here are the results:
The full evaluation framework I used is available on GitHub, and the full results are available here. I did this testing because I was curious how Second Brain would stack up against other agents. This provides the real results, but with a few small caveats, which you can read in the links provided.
NewDemo.mp4
Second Brain runs on your own machine. Two things to install: the app (required, ~2 minutes) and the UI (optional, a ChatGPT-style web app you can add to your phone's home screen).
You need Python 3.11+ and git.
git clone https://github.com/henrydaum/second-brain
cd second-brain
python -m venv .venvActivate the virtual environment:
| Windows | .venv\Scripts\activate |
| macOS / Linux | source .venv/bin/activate |
Then install and run:
pip install -r requirements.txt
python main.pyThat's it — you're in the REPL. requirements.txt is nearly pure Python.
There's also a Dockerfile, if you'd rather not install anything:
docker build -t second-brain .
docker run --rm -it --init -v sb-data:/data second-brainSame REPL, with everything kept in the sb-data volume. The path above is
still the better one for day-to-day use — Second Brain is an assistant for
your machine, and a container starts out unable to see it. Reach for this
when the machine isn't one you want to install on: a server or a NAS, or a
reproducible Linux to test against. docs/DOCKER.md covers it,
including Docker itself if this is your first time.
/setup
The wizard walks you through everything in one pass:
- Install the
essentialsbundle — an LLM backend, file read/edit/search, shell and script running, SQL, web search, subagents, and the Telegram frontend. - Connect a model — paste an API key. Atlas Cloud is the sponsored fast path (300+ models behind one key), but any provider works.
- Telegram (optional) — chat with your Second Brain from your phone. Needs a bot token from @BotFather and your user ID from @userinfobot.
- Web UI (optional) — installs the HTTP frontend and generates your API token, then prints the exact steps to set the app up. More below.
Say hello. You now have a working assistant. If you have questions, just ask.
The REPL works, but it isn't where you want to live. Two much nicer options, in order of effort:
| Effort | What it's like | |
|---|---|---|
| Telegram | ~5 minutes — /setup shows the way |
Push notifications, attachments, inline buttons, and available on all major platforms. |
| Web UI | ~10 minutes | A ChatGPT-style app you open in a browser or add to your phone's home screen. Set it up below. |
Skipped one during /setup? Just run /setup again.
A fresh install is deliberately small. Add capabilities whenever you want:
/packages install
On the Web UI, you can do this in Settings. That opens a picker — browse by category and choose. The bundles worth knowing:
| Bundle | What you get |
|---|---|
bundle_knowledgebase |
Index and search your own files — PDF, Office, images, audio, video, spreadsheets, archives. OCR, transcription, embeddings, and three search tools. Large download. |
bundle_memory |
Durable memory that maintains itself. Notes and skills are surfaced when relevant and written down in the background, as plain markdown you can edit. |
bundle_gmail |
Read, send, reply, label. |
Then tell it which folders to watch:
/config
Set sync_directories to the folders you want indexed. Expect a flood of task messages while the first sync runs — that's normal, and it stops when it finishes.
A modern React frontend built on assistant-ui — a familiar chat interface you can add to your phone's home screen. It lives in its own repo: second-brain-ui.
Works on Windows, macOS and Linux. Takes about ten minutes.
Before you start: Second Brain has to be running while you use the UI — the UI is just a face for it. Leave it going in its terminal and open a second terminal for everything below. You'll also need Node 20.19+ or 22.12+.
The UI proves it's allowed to talk to Second Brain with a token. You need one before anything else.
If you said yes to the web UI during /setup, you already have it — the wizard printed it. Copy it and go to step 2.
Otherwise, in the Second Brain REPL:
/packages install frontend_http
Then run /config, find secret_http_token, and set it to any long random string — mash the keyboard, it just has to be hard to guess. Copy what you set. Then restart Second Brain so the frontend comes online:
/restart
In your second terminal:
git clone https://github.com/henrydaum/second-brain-ui
cd second-brain-ui
npm installNow make your own config file from the example:
| Windows | copy .env.example .env.local |
| macOS / Linux | cp .env.example .env.local |
Open .env.local in any editor. Find the line that reads VITE_SB_TOKEN= and paste your token right after the =, with no quotes and no spaces:
VITE_SB_TOKEN=the-token-you-copied
Leave the other two lines (VITE_SB_URL, VITE_SB_THREAD) exactly as they are — the defaults are correct. Save and close.
npm run devOpen http://localhost:5173 in your browser. You should see your Second Brain, ready to chat.
If you get a blank screen or a 401, the token in .env.local doesn't match the one in /config — that's almost always the problem. Fix it and restart npm run dev.
Start the dev server so it accepts connections from other devices on your network:
npm run dev -- --hostThat prints a second URL (a 192.168.x.x address). To reach it from anywhere rather than just your home Wi-Fi, install Tailscale on both your computer and your phone, and use your machine's Tailscale address instead.
Open that URL in your phone's browser, then add it to your home screen — on iPhone, press the three dots, then Share, and scroll down to Add to Home Screen. Click it, and you're done. It's like a real app from there.
Because it's the only thing that works on every platform today. npm run build produces a dist folder that frontend_http can serve directly (the http_static_dir setting), but a production build deliberately ships no API token — so a browser pointed at it is refused before the page even loads. Serving a build needs a reverse proxy that adds the token, which today exists only in the UI repo's macOS deployment. The dev server is perfectly fine for personal use.
Second Brain creates its data directory on first run. Config, database, installed packages, and the agent's workspace all live there:
| Windows | %LOCALAPPDATA%\Second Brain\ |
| macOS | ~/Library/Application Support/Second Brain/ |
| Linux | ~/.local/share/Second Brain/ |
Run /locations to see the paths on your machine. Useful commands once you're up:
| Command | What it does |
|---|---|
/packages |
Install and remove capabilities |
/config |
Every setting, including plugin settings |
/llm |
Add, switch, or edit model profiles |
/conversations |
Browse and switch conversations |
/commands |
List everything available |
Second Brain is inspired by the human brain. Explorations into neurons turned into the creation of artificial neural networks, which then paved the way for attention mechanisms and transformers. From there came LLMs, and then came the agentic abilities: RAG, tool calls, and cron jobs. With each iteration, Second Brain became closer to its biological inspiration.
Second Brain is still pretty far from the real brain, in many ways. However, it can also do many things better than the human brain ever could. Building it has helped me to better understand the role of AI in my life, and in society. I found the process of building to be extremely valuable, because I realized that the value of AI is that it can be built into so many things. The role of the person is to guide it into productive and creative areas.
MIT
An agent by Henry Daum