brAIn — an AI that manages HA, learns your house, and tells you what’s broken (WIP)
Yo, Ben here. I’ve been building an HA App called brAIn and it’s at the point it’s ready enough for others to try. Fair warning — still a work in progress.
Install: add https://github.com/bruhautomation/BRUH-HA-Apps as a repository under Settings → Add-ons → Add-on Store → ⋮ → Repositories, then install brAIn and sign in from the panel.
Docs: bruhautomation.com/brain
The short version: it runs Claude Code inside Home Assistant with a full set of HA tools, so it can actually manage your instance — not just answer questions about it. It also generates AI insights in an ingress panel, and those cards can go straight onto your Lovelace dashboards.
What it can do
- Fully manage your HA instance — entities, devices, areas, floors, labels, helpers, zones, people, dashboards, blueprints, integrations. Read it, change it, create it, delete it. Plus a real shell in
/configfor everything that’s still YAML, with validation and reloads. - AI-generated insights — ask a question in the panel and get back a card with a real chart, built for your house rather than shipped as a default. Keep the good ones as recurring reports, and drop any of them on a Lovelace dashboard.
- Finds what’s broken — dead batteries, sensors that quietly stopped reporting, devices stuck unavailable, automations that can never fire. Press Fix it and it makes the change; press Not a problem and it never raises that one again.
- Remembers your house — one plain memory document of durable facts (nicknames, household rhythms, the devices that are meant to behave oddly), learned over time and read by every part of brAIn. You can open it, edit it and correct it — it’s not a black box.
- Three ways to talk to it — a chat in the panel that works properly on a phone, the real Claude Code CLI in a browser terminal, or as a conversation agent in Assist so you can just ask out loud. Same house, same memory, whichever one you use.
- Voice can be scoped — the conversation agent has its own access setting. Leave it on “Home Assistant only” and voice can reach entities, services and the registry but not the shell or your files, so a housemate asking Assist a question can’t end up editing your config. Switch it to “Full” when you want it to.
- Callable from automations — services to send a prompt, run a task, run an insight, add to memory or kick off a study session. Your house can ask for help before you notice anything’s wrong.
- Undo — every file is snapshotted before Claude edits it, and
brain undoputs it back. (Still take a real HA backup first — snapshots are not backups.) - CLI tools —
brainandhacommands for memory, study, undo, diagnostics, logs, reloads, entities and services.
Runs on a Claude Pro or Max subscription, or your own API key. One install, one sidebar panel, one login.
Built on other people’s work
The registry-management side is adapted from Spook by Frenck (MIT) — all the “why can’t I do this from a service call” stuff exists because he did it first. And the web terminal started from Claude Terminal by Tom Cassady, which is what showed Claude Code could live inside HA behind ingress at all. Thank you both.
Please remember it’s a WIP
It’s early, it’s opinionated, and some of it will be wrong for your setup. LLMs make mistakes and this one has write access to your config — that’s the trade, and it’s why the undo exists. Take a Home Assistant backup and copy it off the device before you install.
Feedback of any kind is genuinely appreciated — what’s confusing, what broke, what you expected and didn’t get. Reply here, open an issue for bugs, or start a discussion for questions, ideas and setups.
Not affiliated with or endorsed by Anthropic or the Open Home Foundation. brAIn runs the official Claude Code CLI under your own Anthropic account.