A personal AI assistant that runs entirely on your own machine — your knowledge, your AI models, and your credentials stay on your hardware, encrypted at rest under a passphrase only you hold. Your data never leaves your hardware unless you explicitly choose otherwise — and there are exactly two ways to choose:
- Cloud models (optional). Run fully local models (MLX/Ollama) and every prompt stays on-box — the suggested path whenever your hardware can carry one. Add a cloud provider key instead and what you send that model (including knowledge content it is given) goes to that provider — your choice, per model.
- Remote phone access (optional). If you turn it on, your Desktop connects to a
content-blind signaling node to set up an end-to-end-encrypted link to your
phone (the SecureCloudGroup-hosted
rtc.securecloudgroup.comby default, or your own viaSMARTBRAIN_SIGNALING_URL). It sees the connection setup — your two devices' addresses, and when they connect — and never your data, which is encrypted end to end; it keeps no log of who connected.
From one command to your unlocked Chat.
SmartBrain_3000 is a fully local, single-user AI assistant that runs on your own computer — macOS, Windows, or Linux. The launcher assembles what it needs and runs it directly, with no Docker and nothing else to install first — as a desktop tray app, or headless under systemd on a Linux server. Containers remain a first-class alternative wherever you prefer them.
- Your choice of AI. Bring your own API keys for OpenAI, Anthropic, or Google — or run fully local models with Ollama or Apple MLX. A built-in gateway routes between them.
- Your data, encrypted on-device. Your knowledge base, notes, plans, and secrets live in a local database, encrypted at rest under a passphrase only you hold. Real documents are welcome — a several-hundred-page PDF is fine.
- Chat with tools, under your control. The assistant can search your knowledge, track tasks, and act on your behalf — but anything that changes data or reaches out parks for your approval first, and every attempt is audited.
- Talk to it. Tap the mic (or hold Space) and dictate; have replies read aloud; or go fully hands-free in Conversation mode, with a wake word if you want one. Speech is transcribed on your own machine by a built-in engine — nothing to install, nothing sent anywhere. See Voice.
- Follow websites. Subscribe to an RSS/Atom feed and it becomes a knowledge vault that fills itself; tag or delete many items at once. See Feeds.
- It quietly improves itself — within hard bounds. Opt in and SmartBrain reviews its own performance on the cadence you choose (default: every 8 hours) — locally, reversibly, one change at a time — learns your preferences, suggests routines worth automating, and tells you about every change it makes. See the docs on Self-improvement.
- Group and share knowledge with Vaults. Bundle documents into a vault to scope a search, share it as an encrypted file, or publish it publicly so others can subscribe by URL and stay up to date — signed, verified, and re-encrypted under each reader's own passphrase. See Share knowledge with Vaults.
SmartBrain itself needs nothing else — no Docker, no Python, no accounts. On macOS and Windows the install goes through a package manager (Homebrew on macOS, Scoop on Windows); if you don't already have one, install it first with its own one-liner:
- macOS — Homebrew (if
brewreports "command not found"), from brew.sh:/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)" - Windows — Scoop (if
scoopisn't recognised), from scoop.sh, in PowerShell:Invoke-RestMethod -Uri https://get.scoop.sh | Invoke-Expression
Then install the SmartBrain app from https://smartbrain.securecloudgroup.com, or with one command:
macOS — in the Terminal app:
brew install --cask securecloudgroup/tap/smartbrainWindows — in Terminal or PowerShell, via Scoop:
scoop bucket add securecloudgroup https://github.com/SecureCloudGroup/scoop-bucket
scoop install securecloudgroup/smartbrainLinux (x86_64) — download the install script, read it, run it. It verifies the
release's signature and checksum, installs per-user (no root), and adds SmartBrain to
your app menu; --headless sets up a systemd --user service for a server instead:
curl -fsSLO https://raw.githubusercontent.com/SecureCloudGroup/SmartBrain_3000/main/installer/install-linux.sh
sh install-linux.sh # desktop; or: sh install-linux.sh --headlessThen smartbrain start — if that says command not found, log out and back in
(your shell picks up ~/.local/bin once it exists). Prefer containers (or on arm/musl, where there is no native
build yet)? The Docker stack is one file:
curl -fsSLO https://raw.githubusercontent.com/SecureCloudGroup/SmartBrain_3000/main/compose/docker-compose.release.yml
docker compose -f docker-compose.release.yml up -dThen open http://localhost:33000.
Homebrew and Scoop install a small desktop launcher — a menu-bar / tray app. On its first start it downloads and verifies everything SmartBrain needs (a Python runtime, the app itself, and the model gateway) into a folder it owns, then runs it and opens your browser. That download is a few hundred megabytes, so allow a few minutes the first time; after that it starts in seconds. The app runs at http://localhost:33000. In the app:
- Set a passphrase and save your Emergency Kit (a one-time Recovery Key). There is no password reset — the Recovery Key is the only way back in if you forget your passphrase, so store it somewhere safe and offline.
- Connect a model under Settings: add a cloud provider API key, or run a local model — MLX on an Apple-Silicon Mac, Ollama on any OS, or either one running on another machine you own (SmartBrain on a Linux box, models on a Mac's GPU — see the models guide). If one is already running locally, SmartBrain detects it and offers a one-tap connect right on the Chat screen.
- Start chatting.
Does it need Docker? Not on an Apple-Silicon Mac, 64-bit Windows, or x86_64 Linux — those run it directly. Intel Macs install the same desktop app but fall back to Docker, because there is no native build for them; arm/musl Linux runs the Docker stack above.
SMARTBRAIN_NATIVE=0forces the Docker path on any machine.
Building from source? Contributors can
git clonethe repo and runpython3 installer/install.py install(this needs Docker, git, and Python, and compiles the image locally). See Install from source.
Short, silent clips (~15s each) for every step — the five Quickstart clips take you from install to fully working; the rest are optional power-ups:
SmartBrain updates itself — there is no command to run. The launcher looks for a new
release in the background and downloads it without disturbing what you're doing. Once it's
ready, the app itself says so and offers Install now; the same update is offered in the
menu as Install update now / Install on next start. Installing restarts SmartBrain
(under a minute), so you unlock again afterwards. Ignore the notice and the update installs
the next time you start. The desktop app updates itself the same way, so brew upgrade and
scoop update are no longer part of normal use. Don't want to wait for the background check?
Check for updates in the menu looks right now. Installing keeps the previous version on
disk for rollback and prunes anything older; a download that fails says so in the menu.
- Which version am I on? The menu names it, and so does the app under the logo, top-left — and Settings → Status shows it alongside the lock state, model server, voice engine, schedules, feeds and paired devices, which is the first place to look when something is off.
- From a paired phone you can see that an update is waiting, but installing it is Desktop-only.
- Linux desktop launchers relaunch themselves after an update, so the tray icon comes back on its own.
- Headless Linux swaps under systemd and the unit's
Restart=brings up the new version — nothing to run there either. - On the Linux Docker stack: no launcher does it for you — re-run
docker compose -f docker-compose.release.yml pullthendocker compose -f docker-compose.release.yml up -d. - From source:
python3 installer/install.py updatebacks up your encrypted data first, then rebuilds, restarts, and verifies — prompting before changes, on the host, never inside the container.
An update never touches your data. It lives in a folder you own
(~/Library/Application Support/SmartBrain/data on macOS, ~/.local/share/smartbrain/data
on Linux); the Linux Docker stack keeps it in named Docker volumes.
These are advanced tiers — none are needed for the Quickstart above:
- Gmail — connect a Gmail account (your own Google OAuth client, loopback flow) so the assistant can read and draft mail. See the docs.
- Your phone, from anywhere — pair a phone to reach your assistant on Wi-Fi or cellular over an end-to-end-encrypted WebRTC link, with no router port-forwarding. Off by default. See Remote access.
- Your knowledge in another AI client — SmartBrain is also an MCP server, so Claude Desktop, Cursor, or any MCP client can search and read your knowledge base (read-only, loopback, behind a token you mint). Off until you generate that token. See Connect external tools.
Full guide: docs/ (also available in-app under Help).
SmartBrain_3000 is licensed under the Elastic License 2.0 (ELv2). It is source-available, not OSI "open source." In plain terms:
- ✅ You may use, self-host, and modify it for free — including inside a business, for your own use.
- ❌ You may not offer it to others as a hosted or managed service, resell it, or ship a competing product built from it.
See LICENSE for the full terms.
More: Changelog · Contributing · Security policy · Ask a question
© 2026 The Frels Holdings LLC. "SmartBrain", "SmartBrain_3000", and "SmartBrain AI" are trademarks of The Frels Holdings LLC.
Outside contributions are credited here and in the changelog entry they ship in. Every open
issue carries a Done means line; good first issue means you can finish it from the issue
text alone, help wanted means it's welcome and may need a short conversation first, and
needs design means a maintainer decision is still pending — comment if you want to shape it.
Start at Where help is wanted,
which also carries the roadmap.
- (your name here — the first outside PR starts this list)
Stuck, or something didn't work the way this page said it would? Ask in Discussions — questions about installing, connecting a model, or getting your documents in are all welcome, and no question is too basic. Found a reproducible bug? Open an issue.
A word on response times: SmartBrain_3000 is built and maintained part-time. Everything posted here gets read, and issues are addressed as quickly as possible — data-safety problems first, broken installs next, everything else after that — but there is no guaranteed turnaround. The best way to speed a fix along is a clear report: what you did, what you expected, what happened instead, and the relevant lines from Open logs in the menu.
Please report security issues privately — see SECURITY.md
(contact: info@securecloudgroup.com). Do not open public issues for
vulnerabilities.