Give your
desk a voice.

One command. A visible response.

Choose a simulated command. This demo requests no microphone access and controls no real devices.

Simulation ready. No Home Assistant connection.

Build the system, not just the speaker.

Reference: Home Assistant documentation displaying version 2026.10.4, checked October 31, 2026. The S3-BOX guide states a 2023.12 minimum; use a currently supported installation. The web installer selects its maintained firmware; record the firmware version installed on your device.

Where your command goes.

On-device wake-word detection is only one stage. Speech processing and device integrations determine whether the complete command path stays local.

Get a first win in five stages.

1. Prove the lamp works. In Home Assistant, turn your chosen light on and off using the dashboard. Give it a clear name such as Desk Lamp. Resolve integration problems before adding voice.
2. Prepare Assist. Choose local or cloud speech processing below. Test the typed command “turn on Desk Lamp” in Assist. A successful text test isolates entity names and control from microphone problems.
3. Flash the exact box variant. Open the linked official S3-BOX guide in desktop Chrome/Edge. Connect USB-C directly to the box, not its dock. Follow its variant-specific install, Wi-Fi and Home Assistant onboarding prompts. The installer replaces existing device firmware.
4. Select and test the voice pipeline. Open the discovered ESPHome device, select your configured assistant as described in the official guide, and check its wake-word setting. Say the configured wake phrase, then “turn on Desk Lamp.” Do not assume a particular wake phrase is enabled.
5. Test repeatedly. Try on/off commands from your normal desk position. Record successful attempts, misunderstood commands and response times. Test microphone mute, then repeat after restarting the satellite.

Choose your voice-processing route.

For the local route, install and start your chosen speech-to-text app plus Piper on HAOS. Add their discovered Wyoming integrations. In Settings → Voice assistants, create an assistant using Home Assistant as conversation agent, your selected speech engine, Piper, and matching language settings.

Expose Desk Lamp under Settings → Voice assistants → Expose. Start with that single entity. A locally processed command may still operate a cloud-dependent light integration; check the entire chain before describing your system as offline.

Upgrade: “start desk mode.”

Once the basic command works, create an automation with a Sentence trigger for start desk mode. Add a light-on action targeting your lamp and a conversation response. Test the typed sentence before trying it aloud.

Compatibility: use a Home Assistant conversation agent. With Speech-to-Phrase, verify the custom sentence is supported by the installed model/version; typed success alone does not establish that it can transcribe an arbitrary phrase. Use an open-ended speech engine if needed.

alias: WolfieWeb - start desk mode
description: Example only - replace the light entity before enabling
triggers:
  - trigger: conversation
    command:
      - start desk mode
conditions: []
actions:
  - action: light.turn_on
    target:
      entity_id: light.replace_with_your_desk_light
  - set_conversation_response: "Desk mode is ready."
mode: single

Paste this into a new automation’s YAML editor, replace light.replace_with_your_desk_light with the actual entity ID, and save. This is an untested starter example; it is not an ESPHome device configuration.

Troubleshoot the stage that failed.

Keep the audio satellite focused on voice; ESPHome warns that extra components, especially BLE alongside audio, can cause resource problems. There is no need to add an LLM to switch a desk lamp.

Study links

Verify board stock and firmware compatibility before purchasing. No price, recognition accuracy or response-time claim is made here. Hardware validation remains outstanding.