Control plane
Central authority for APIs, operator UI, governance, trust, scheduling, messaging, and system-wide decisions.
Local-first voice infrastructure
An AI system
you can touch.
Hexe is a modular, local-first control plane made tangible through custom voice endpoints—from governance and runtime control to wake word, screen, speaker, and response.
01 / SYSTEM
02 / WHAT HEXE IS
A control plane
with a voice.
Hexe brings home, edge, and trusted external execution under one structure. Core holds authority and policy; Supervisor owns host-local runtime control; Nodes declare capabilities and execute across explicit trust boundaries.
Voice is where that architecture becomes physical. Microphones, displays, transport, speech services, and interaction logic work as one inspectable stack—without treating the room as a cloud peripheral.
It is an active open-source engineering project: working software and physical devices first, with the edges documented honestly.
Host runtime
Local authority for system health, resource visibility, lifecycle control, and workload management.
External execution
Trusted systems declare capabilities, receive governance, publish telemetry, and execute beyond Core.
03 / SIGNAL PATH
From sound to intent
and back again.
Inside the wider Core → Supervisor → Nodes platform, a real-time WebSocket path carries commands and PCM audio between room endpoints and HexeVoice. HexeInteraction turns recognized speech and shared events into useful behavior.
EDGE / 01
ESP32 endpoints
Wake word · microphones · speakers · touch displays
Physical edge
PCM + COMMANDS
VOICE / 02
HexeVoice
Streaming · STT · TTS · speaker identity · device control
Audio runtime
EVENTS + INTENT
LOGIC / 03
HexeInteraction
Context · automations · task routing · response behavior
Interaction layer
04 / CURRENT CAPABILITIES
Built, running,
and being refined.
Working now Active development
WORKING NOW01
Local wake + streaming audio
Endpoints detect a wake phrase, then move command and PCM traffic over a persistent WebSocket architecture.
WORKING NOW02
Aa
Speech in, speech out
STT and TTS are part of the same system path, with local options and ready-state audio handling.
WORKING NOW03
◎
Speaker identification
Hexe can associate voices with enrolled speakers while keeping identity handling grounded in the local system.
ACTIVE DEVELOPMENT04
∿
Ambient-noise awareness
Calibration and room-aware signal behavior help endpoints understand the acoustic space around them.
WORKING NOW05
⌂
Multiple endpoints
Room devices register independently, report state, and receive targeted screens, audio, and commands.
WORKING NOW06
↥
Over-the-air updates
Firmware and interface assets can move through managed update paths without dismantling an endpoint.
05 / HARDWARE
Designed past
the breadboard.
Hexe runs across custom endpoint builds, from compact ESP32-S3 voice devices to a Waveshare ESP32-P4 touch display with a full visual surface. Firmware, audio, UI assets, provisioning, and OTA all meet at the edge.
- ESP32-S3Voice-focused endpoints
- ESP32-P41024px touch display class
- DISPLAY UIClock, weather, timers, notices
- DEVICE OPSBLE onboarding + OTA
HEXE / KITCHEN● ONLINE
07:42
Tuesday · 68° · Clear
LISTENINGWhat can I do?
ENDPOINT / P4-7BTouch + audio + local state
06 / TWO SPECIALISTS
V
AUDIO RUNTIME
HexeVoice
The real-time layer between microphones, speakers, displays, and speech services. It owns endpoint connections, audio transport, voice processing, and device-facing presentation.
- WebSocket command + PCM transport
- Wake word, STT, TTS, speaker ID
- Firmware endpoints, screens, OTA
I
BEHAVIOR LAYER
HexeInteraction
The context and coordination layer. It turns recognized speech, timers, weather state, and shared events into responses that make sense for a person, room, and endpoint.
- Intent and task coordination
- Event-driven automations
- Contextual response behavior
07 / ROADMAP
What comes next.
Roadmap items are direction, not shipped capability. Hexe grows by proving each layer on real devices before calling it done.
NOW
Stronger room awareness
Continue refining calibration, noise context, and endpoint-specific behavior.
NEXT
Richer visual interactions
More backend-owned screens and event-driven UI across capable displays.
LATER
Multi-room music
Synchronized playback and room-aware media are a destination—not a current feature.
08 / OPEN SOURCE
Follow the build,
not the pitch.
Hexe is developed in public across the Core, Voice, and Email repositories. Hexe Email can connect to Gmail—with the user’s permission—to fetch, display, classify, summarize, label, and send email under user or operator control. The code is the clearest view of what exists today; the docs preserve the platform-level architecture and operating contracts.