INTRODUCING JANET
Janet is a local-first AI assistant that controls your smart home, understands context instead of commands, and runs entirely on hardware you own. Your data never leaves the house.
THE PROBLEM
Cloud-tethered, command-driven, and built around a wake word you have to remember. The category hasn't moved — Janet does.
Most assistants route every request through remote servers. Your home shouldn't stop working when your internet does.
You shouldn't have to repeat a device's name every time you speak. Real conversations don't work like that.
Today's assistants treat each utterance as an isolated command. They don't track who's in the room, what just happened, or what you actually meant.
Your habits, routines, and voice are processed in the cloud and retained on someone else's terms. Convenience shouldn't cost ownership.
WHAT MAKES JANET DIFFERENT
Local processing, contextual understanding, real privacy, and the hardware to make it real.
Core intelligence — speech, intent, reasoning — runs entirely on a device in your home. The cloud is optional, never required.
Janet uses context — who's in the room, what just happened, how you're speaking — to know when you're talking to it. No "Hey Janet" required.
Lights, locks, climate, scenes — controlled locally. Your routines, schedules, and voice never leave the house.
Supporting capabilities
Tracks conversation history, environment, and prior intent so follow-ups work the way you'd actually phrase them.
Filters background noise, detects speech, identifies who's speaking, and transcribes — all locally, with no raw audio retained.
Event-driven visual understanding for advanced features. Off by default, on only when you want it.
Optimized for NVIDIA Jetson-class hardware so the AI core runs at conversational speed without a datacenter.
HOW IT WORKS
Audio in. Intent out. Reasoning in the middle. Vision when you want it.
AUDIO
Janet keeps a short rolling buffer of sound — never stored, never uploaded. It detects speech, filters out music and TV, identifies who's speaking, and turns it into text. The raw audio is gone the moment the transcript is ready.
Listen
Detect
Identify
Output
ROUTING
Janet weighs what was said, who said it, what just happened, and where you're standing. If it's for Janet, the request gets routed to the right system. If not, it stays quiet.
REASONING
A layered system: a quick intent classifier for everyday requests, a real-time action processor for smart-home commands, and a reasoning engine for the questions that actually need thinking. Simple stays fast. Complex goes deeper.
Intent Classification
Fast routing for simple, recognised tasks.
Low-latency Processors
Real-time device control and action execution.
Local LLM Engine
Deep reasoning for complex, open-ended queries.
VISION
Vision is opt-in and event-driven. A low-power awareness mode runs only if you enable it; heavier visual processing fires on specific triggers, not continuously. Off is the default.
THE ROOM
Janet listens through every microphone you set up — kitchen, living room, bedroom. Wherever you are, it hears you. The signals converge on a single on-device intelligence.
PRIVACY
Local processing isn't a marketing word for us — it's the architecture. Here's what that actually means.
Speech, intent, and reasoning run on hardware in your home.
Audio is processed in a rolling buffer and discarded — only transcripts persist.
Cloud is something you opt into per feature, not a requirement to function.
The hardware is yours. The data is yours. The model weights live with the device.
| JANET | Typical | |
|---|---|---|
| Core processing | Runs locally | Runs in the cloud |
| Wake word | Not required | Required |
| Raw audio | Discarded after processing | Often stored or processed remotely |
| Data ownership | You control it | The platform controls it |
| Extensibility | Open by direction | Closed ecosystem |
WHY TRUST JANET
Janet is shipping in public — every decision, every commit, every dead end. You don't have to take our word on privacy or local-first. You can read the code.
The source is public. The architecture is auditable. Nothing about Janet's privacy story is on faith.
Designed from the start to run on hardware you own, not adapted from a cloud product after the fact.
Roadmap, design notes, and the reasoning behind each system live alongside the code.
Real local inference on real local silicon — Jetson-class edge hardware, not a Raspberry Pi pretending.
Follow Janet as it evolves from concept to a real local AI system.
