| Google Cloud project | For Actions Console and Dialogflow |
| Public URL | ngrok (testing) or domain with HTTPS (production) |
Step 1: Set Up Public Access
# For testing: use ngrok to expose your local OpenClaw
ngrok http 3000
# → Forwarding: https://abc123.ngrok.io → http://localhost:3000
# For production: configure your reverse proxy
# Your OpenClaw endpoint will be:
# https://your-domain.com/api/google-assistant/webhook
Step 2: Install the Google Assistant Skill
openclaw skill install google-assistant
openclaw skill setup google-assistant
# → Public webhook URL: https://your-domain.com/api/google-assistant/webhook
# → Invocation name: open claw
# → Language: English
# → Configured!
Step 3: Create a Google Actions Project
- Go to Google Actions Console
- Click New project → Name it "OpenClaw Assistant"
- Choose Custom → Blank project
- Go to Develop → Invocation → Set display name to "Open Claw"
Step 4: Configure Dialogflow (or Actions Builder)
// Webhook configuration
{
"webhook": {
"url": "https://your-domain.com/api/google-assistant/webhook",
"headers": {
"Authorization": "Bearer ${GOOGLE_WEBHOOK_SECRET}"
}
}
}
Step 5: Define Intents
Set up a catch-all intent that forwards everything to OpenClaw:
{
"intent": "openclaw.query",
"trainingPhrases": [
"Ask OpenClaw *",
"Tell OpenClaw *",
"OpenClaw *"
],
"fulfillment": {
"webhook": true
}
}
OpenClaw Configuration
# ~/.openclaw/skills/google-assistant/config.yaml
google_assistant:
webhook:
path: /api/google-assistant/webhook
secret: ${GOOGLE_WEBHOOK_SECRET}
voice:
response_format: ssml # ssml or plain
max_response_length: 640 # Characters (Google limit)
speaking_rate: medium
behavior:
greeting: "Hi! How can I help?"
error_message: "Sorry, I couldn't process that. Try again."
timeout_message: "I'm taking too long. Let me try a simpler answer."
permissions:
location: true # Access user's location
name: true # Access user's name
responses:
include_suggestions: true # Show suggestion chips on screens
max_suggestions: 3
Voice Interaction Examples
"Hey Google, talk to Open Claw"
Google: "Hi! How can I help?"
"What's on my calendar today?"
Google: "You have 3 events: A standup at 9, a product review at 11,
and a 1-on-1 at 2 PM."
"Set a reminder to buy groceries at 5 PM"
Google: "Done! I've set a reminder for 5 PM to buy groceries."
"Send a message to Alex saying I'll be late"
Google: "I'll send Alex a message saying you'll be late. Should I
send it now?"
"Yes"
Google: "Message sent to Alex."
SSML Response Formatting
OpenClaw can send rich voice responses using SSML:
<speak>
You have <say-as interpret-as="cardinal">3</say-as> meetings today.
<break time="500ms"/>
The first one is at <say-as interpret-as="time">09:00</say-as>,
a standup meeting.
</speak>
Multi-Device Support
| Device | Features | Notes |
| Google Home / Nest | Voice only | Keep responses concise |
| Nest Hub (screen) | Voice + visual | Suggestion chips, cards |
| Android phone | Voice + visual + touch | Full interaction |
| Smart displays | Voice + visual | Rich card layouts |
Testing
# Test the webhook locally
curl -X POST https://your-domain.com/api/google-assistant/webhook \
-H "Authorization: Bearer $GOOGLE_WEBHOOK_SECRET" \
-H "Content-Type: application/json" \
-d '{
"handler": {"name": "openclaw.query"},
"intent": {"query": "What is the weather today?"}
}'
# Test in Actions Console
# Go to Test tab → Type or speak your query
Troubleshooting
| Problem | Solution |
| "I don't know how to help" | Check invocation name matches config |
| Webhook timeout | Reduce response length or use a faster model |
| No response from OpenClaw | Verify gateway is running and publicly accessible |
| "Sorry, this action is not available" | Deploy the action in Actions Console |