This one works for you.
It isn't a website. It isn't an account.
It's a presence in your home that knows your voice.
You said it out loud.
It answered out loud —
and it started speaking
before it finished thinking.
Replies stream sentence-by-sentence. First words in moments, not after the whole thought.
what changes
Hands full, phone across the kitchen.
You just talk.
The thought that won't let you sleep.
Say it. Let it go.
Your kid asks it why the sky is blue.
The question never leaves your house.
Now — the part nobody believes.
The cloud was never invited.
The microphones were cheap. The models were ready.
But a private assistant can't be sold twice —
and you were the second sale.
So the industry built smart speakers that answer to headquarters.
Inhouse is the one with no headquarters.
alright — here's how
Any Linux box — a mini-PC, a home server, a small VPS. Two CPU cores are genuinely enough.
Run a local model with Ollama and unplug the internet — or point it at Claude, GPT, or any API. Swappable in one line.
Whisper listens locally. Piper speaks locally. Your audio never becomes someone's training data.
An installable app on your phone, reached through an encrypted tunnel only your devices can see. Interrupt it mid-sentence — it yields.
# five minutes from now, it says hello
git clone https://github.com/getinhouse/inhouse
cd inhouse && make setup && make voice
.venv/bin/python -m inhouseMIT licensed · 60 tests · no telemetry, no account, no asterisk
The core is complete and stays open. Pro is the production playbook —
for the day it stops being a project and starts being part of the house.