Hardware ESP32-S3

Syntax.
Your Wrist. Your Desk. Your AI.

ReSono Labs Syntax is a two-device AI voice terminal ecosystem. A touchscreen watch on your wrist and a companion terminal on your desk — both powered by the same OpenClaw brain, both speaking with one voice.

The Hardware

Two Devices. One Brain.

⌚

Syntax Watch

A 2.06" AMOLED touchscreen AI companion on your wrist. Always on, always listening for your next command — with a notification drawer for delegated AI task results.

Watch AMOLED Voice AI
Details below ↓
🖥

Syntax

A 1.85" round-display desktop companion terminal. Sleek, always-plugged-in, always-connected. The desk anchor for your ambient AI workflow.

Desktop Round LCD Voice AI
Details below ↓
ReSono Labs Syntax Watch and Desktop Companion
Unified Design Language

Both Devices, One Ecosystem

Syntax Watch and Syntax share the same UI framework, the same OpenClaw voice protocol, and the same notification inbox. Start a conversation on the desktop, walk away, and continue it on your wrist. Or delegate a heavy task from the watch and get the spoken result when you're back at your desk.

They look different because they are different hardware. But underneath the shell, they're the same device — running the same agnostic ESP32-S3 firmware.

Architecture

Thin Terminal. Zero AI on Device.

Neither the watch nor the desktop terminal runs an LLM. They're thin clients — they capture your voice, stream it to OpenClaw over WebSocket, receive the AI's audio response, and play it back. The heavy lifting (transcription, inference, text-to-speech) all happens on the server.

This design means no AI API keys ever touch the device. No credentials in firmware, no secrets in NVS. The ESP32 just streams PCM audio and paints pixels. Everything sensitive lives in OpenClaw.

Switching from Gemini Live to OpenAI Realtime takes one CLI command: openclaw config set liveProvider google

SYNTAX DEVICE (ESP32-S3)
Audio capture → PCM stream → WebSocket
↓ WebSocket (TLS)
OPENCLAW GATEWAY
Session routing, auth, tool execution
↓
Gemini Live
OpenAI Realtime
Firmware

One Firmware. Any ESP32-S3 Hardware.

The agnostic architecture uses driver tables (ops structs) — the firmware doesn't hardcode any specific screen, speaker, or codec. It dispatches through function pointers defined per board profile. Add a new ESP32-S3 form factor? Implement the driver table. Everything else — the AI, the UI framework, the voicemail system — works out of the box.

📺

Display Driver Table

Init, flush, sleep, wake, brightness — all dispatched through board-specific function pointers. Works with rectangular AMOLED or round LCD.

🎤

Audio Driver Table

I2S mic start/stop, speaker PCM write — via ES8311/ES7210 codecs with automatic 24k→16k resampling for the AI stream.

👆

Input Driver Table

Touch, tap, and gesture events routed through a board-specific input handler — whatever input hardware the device has.

🔋

Power Driver Table

Battery monitoring, charging state, AXP2101 PMU integration — per-board power profile with shared battery logic.

🌐

Provisioning Portal

Captive portal for initial Wi-Fi setup. Credentials never hardcoded — securely provisioned once, stored in NVS.

🔄

LVGL UI Framework

Custom layered LVGL interface with a top System Drawer, bottom Results Drawer, and a center Orb Service visualizer.

You: "Research this codebase for me"
→ Delegated to background worker
Watch notification: "3 results ready" ↑
Tap notification: Gemini briefs on result → speaks summary
Voicemail System

Delegate. Walk Away. Get Spoken Results.

When you ask Syntax to do something complex — research a codebase, draft a report, analyze a codebase — it doesn't block the conversation. It delegates to a background text worker (an OpenClaw subagent) and goes silent.

When the result is ready, the device gets a deskbot.results.ready event. A notification badge appears on the watch or desktop screen. Tap it, and Gemini is briefly briefed on the result and speaks a concise natural-language summary back to you.

The voicemail broker is durable JSON state — tasks survive network blips, device reboots, and context resumptions.

Memory Architecture

Tiered History. Zero Context Loss.

🔥

Hot Active History

Up to 240 recent conversation turns stored in a fast-access JSON file. Always in context, never re-fetched.

📦

Archived History

Older turns automatically moved to archive — keeps active snapshots lightning fast without losing any conversation history.

🔁

OpenClaw Mirroring

Every Syntax conversation mirrors directly into your main OpenClaw agent. Talk to your desk bot, and your primary AI remembers it on your phone.

⚡

Context Resumption

Resumption handle logic sustains context across network interruptions and device reboots — no AI repetition, ever.

Audio Intelligence

Silence Guard. No Clipping. Ever.

The Silence Guard architecture monitors AI and user audio state in real time. If you're speaking or Gemini is mid-sentence, background polling and worker spawns are suspended completely. No audio stuttering. No mid-word interruptions.

After you stop speaking, the system waits exactly 1 second before resuming background work — ensuring no "clipping" at the end of your turn.

The Orb Service visualizes AI state on both devices — its animation speed and color theme track whether the system is Idle, Listening, Thinking, or Speaking.

IDLE — Orb dark, no traffic
LISTENING — Guard active, no spawns
THINKING — Orb pulsing, workers live
TALKING — Orb animated, stream active
+1s SILENCE BUFFER — safe to resume
Security

No Keys on Device. No Exceptions.

Centralized Key Management

All AI API keys live strictly inside OpenClaw. Gemini and OpenAI credentials never reach the ESP32, never touch NVS, never appear in firmware.

Token-Based Device Auth

Each device generates a random 6-digit PIN stored in NVS. That PIN is the bearer token for the local web interface — revocable at any time from OpenClaw.

Session Auto-Expiry

Web sessions expire after 30 minutes of inactivity. PIN re-entry required. Clean kill switch for any lost or compromised device.

Thought Filtering

Built-in filtering ensures Gemini's internal model reasoning (thoughts) never leaks to the user interface. Only final outputs are surfaced.

Specs

Built on ESP32-S3

Both Syntax devices run on the ESP32-S3, chosen for its dual-core Xtensa architecture, Wi-Fi + BLE 5.0 radio, and broad ecosystem support. The firmware is built on ESP-IDF v5.4.2 with LVGL for the UI framework.

ESP32-S3 ESP-IDF v5.4.2 LVGL WebSocket (TLS) I2S Audio ES8311 / ES7210 AXP2101 PMIC SH8601 AMOLED FT5x06 Touch NVS
ReSono Labs Syntax concept render

Want to Build with Syntax?

ReSono Labs Syntax is an open-architecture ESP32-S3 platform. Talk to us about custom hardware form factors, white-label deployments, or integrating Syntax into your own product line.

hello@resonolabs.com Meet the Team