Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

docs(home-satellite): document that the token role runs without building a hub

nirholas6fc8dccff1ec
diff --git a/services/home-satellite/README.md b/services/home-satellite/README.md
index 685dfa129..48d7d415d 100644
--- a/services/home-satellite/README.md
+++ b/services/home-satellite/README.md
@@ -150,6 +150,7 @@ fixture: Home Assistant runs its own speech recognition, its own intent handling
to speech, and the house really changes.
```bash
+# The house's own voice stack: speech to text, text to speech, wake word.
docker network create wyoming
docker run -d --name whisper --network wyoming rhasspy/wyoming-whisper:latest \
--model tiny-int8 --language en --uri tcp://0.0.0.0:10300 --data-dir /data --download-dir /data
@@ -157,24 +158,31 @@ docker run -d --name piper --network wyoming -p 10200:10200 rhasspy/wyoming-pipe
--voice en_US-lessac-low --uri tcp://0.0.0.0:10200 --data-dir /data --download-dir /data
docker run -d --name openwakeword --network wyoming rhasspy/wyoming-openwakeword:latest \
--preload-model ok_nabu --uri tcp://0.0.0.0:10400
-docker run -d --name ha --network wyoming -p 8123:8123 \
- --add-host=host.docker.internal:host-gateway ghcr.io/home-assistant/home-assistant:stable
-# Onboard Home Assistant and get a token (repo root).
-node scripts/provision-home-assistant.mjs --url http://localhost:8123 --rooms
+# Home Assistant, onboarded and seeded, printing its URL and token (from the repo
+# root). It labels the container it creates and refuses to touch one it did not,
+# which matters on a machine running several instances at once.
+node scripts/home-test-instance.mjs --up --onboard --seed --name satellite --json
+# It lands on the default bridge, so join it to the voice stack's network:
+docker network connect wyoming three-ws-home-test-satellite
# Add the four Wyoming services, build an Assist pipeline, make it preferred.
-node scripts/provision-ha-pipeline.mjs --url http://localhost:8123 --token "$HOME_ASSISTANT_TOKEN" \
+# The satellite is addressed at the wyoming network's gateway, which is how a
+# container reaches a service listening on the host that runs it.
+node scripts/provision-ha-pipeline.mjs --url "$HOME_ASSISTANT_URL" --token "$HOME_ASSISTANT_TOKEN" \
--stt whisper:10300 --tts piper:10200 --wake openwakeword:10400 \
- --satellite host.docker.internal:10700
+ --satellite "$(docker network inspect wyoming --format '{{range .IPAM.Config}}{{.Gateway}}{{end}}')":10700
# Speak a sentence into the pipeline through this satellite and print what came back.
node scripts/pipeline-run.mjs \
--viewer ws://127.0.0.1:10701/viewer --token "$(node src/index.js token)" \
- --piper 127.0.0.1:10200 --say "turn off the kitchen lights" \
- --ha http://localhost:8123 --ha-token "$HOME_ASSISTANT_TOKEN" --watch light.kitchen_lights
+ --piper 127.0.0.1:10200 --say "turn on the bed light" \
+ --ha "$HOME_ASSISTANT_URL" --ha-token "$HOME_ASSISTANT_TOKEN" --watch light.bed_light
```
+Run the satellite itself in Docker instead and Home Assistant addresses it by container name
+(`--satellite three-ws-satellite:10700`) with the container on the same `wyoming` network.
+
`pipeline-run.mjs` plays the part of the browser: it connects to the viewer WebSocket exactly as
`src/home/satellite.js` does, streams microphone audio the same way, and receives the same events
and the same speech the avatar lip-syncs to. The "microphone audio" is real speech, synthesized by