This week Pika dropped one of the most sci-fi features in AI tooling so far: real-time video call support for AI agents. Send your OpenClaw agent a Google Meet invite. It joins the call. It participates. It takes notes.
Here's what that actually means in practice and how to get it working on your setup.
What Pika's Video Chat Integration Does
Pika is a real-time multimodal communication layer that sits between OpenClaw and video conferencing tools. When connected to your OpenClaw agent:
- Your agent gets a calendar-connected email address (e.g.,
doit@youragent.pika.ai) - Invite that address to any Google Meet call like you'd invite a human participant
- The agent joins the call, processes audio in real time, and can respond via text-to-speech
- After the call: automatic summary, action items, and decisions pushed to your OpenClaw workspace
Important distinction: The agent doesn't have a face or camera — it joins as an audio-only participant with a name. Other participants see it listed in the call like any other attendee. It can speak when addressed directly or stay in listen/note mode.
Prerequisites
- OpenClaw running and accessible (not just localhost — needs a public URL or ngrok tunnel)
- Pika account (pika.ai — currently free tier available during beta)
- Google account for calendar integration
- OpenClaw gateway with a valid auth token (required for Pika's webhook)
Setup: Step by Step
Go to pika.ai → Sign up → Create Agent. Give it your agent's name (from your IDENTITY.md), description, and connect your OpenClaw gateway URL + auth token. This is where Pika will forward transcripts and receive responses.
In Pika dashboard → Integrations → Google Calendar → Authorize. This gives your agent its own calendar presence. Pika creates a dedicated Google account for the agent that you control.
In your OpenClaw openclaw.json, add the Pika plugin config block with your Pika API key and webhook secret. Restart OpenClaw. Pika will verify the connection with a test ping.
Add a ## Meeting Behavior section to your SOUL.md. Define: active participant or silent note-taker? When to speak? What to do with the transcript after? This controls how your agent behaves in calls.
Create a Google Meet event → add your agent's Pika email as a guest → start the call. Check that the agent joins, that transcription is flowing in your OpenClaw logs, and that the post-call summary appears in your workspace.
What to Put in Your SOUL.md Meeting Section
This is the most important configuration step — and the one most setup guides skip. Your agent's meeting behavior should match how you actually work:
Example SOUL.md meeting block:
## Meeting Behavior
- Default mode: silent note-taker unless directly addressed by name
- When addressed: respond concisely, max 2 sentences unless asked to elaborate
- After every call: generate summary → action items → decisions made → open questions
- Save transcript to memory/meetings/YYYY-MM-DD-[meeting-name].md
- Never speak over humans. Wait for a natural pause.
Real-World Use Cases
Client meetings
Agent takes notes silently. Pushes action items to your workspace immediately after the call ends. No more "who said what" debates.
Property walkthroughs
Walk through a property on video call with a tenant or contractor. Agent captures every maintenance request and commitment made during the call.
Contractor briefings
Brief a contractor on a job verbally via Meet. Agent transcribes the scope, generates a written summary, and files it automatically.
Sales calls
Agent listens to discovery calls and pulls deal notes, objections, and next steps into your CRM or workspace automatically.
Tax/legal consultations
Agent joins your CPA or attorney call and captures decisions, action items, and deadline commitments — nothing slips through the cracks.
Team standups
Agent joins daily standups, logs blockers and commitments, and sends you a summary with anything that needs your attention flagged.
Known Limitations (Beta)
- Google Meet only — Zoom and Teams support are on Pika's roadmap, not live yet
- Text-to-speech latency — there's a 1–3 second delay when the agent speaks, which can feel awkward in fast-moving conversations
- Requires public gateway URL — if your OpenClaw is on localhost only, Pika can't reach it; you need a public domain or an ngrok tunnel
- Audio quality sensitive — poor call audio leads to transcription errors; a headset on your end improves accuracy significantly
- Beta pricing TBD — free during beta, pricing not yet announced for production
The public URL requirement is the most common blocker. If your OpenClaw isn't already accessible via a proper domain with HTTPS, you'll need that in place before Pika works. That's one of the things ClawReady sets up on every install — it's not an afterthought.
Is This Worth Setting Up Now?
If you're already running OpenClaw with a proper public setup, yes — the Pika integration adds genuine value for anyone who does meetings. The note-taking use case alone pays for itself in the first week.
If your OpenClaw is still on a laptop with no public URL, get the base setup solid first. The Pika integration will be meaningless if the foundation isn't there.
For new setups where you want Pika working from day one — mention it on your ClawReady intake call and we'll include the Pika configuration and SOUL.md meeting section as part of the setup.