Draft out loud, edit later.
Talk through a whole opening, an argument, a piece of training material. Land the idea on the highway, read the transcript at the desk, polish the words you actually said.
Push-to-talk voice for thinking out loud into your main OpenClaw session — most often from the car. Tap the big button to toggle talk mode, speak through a whole paragraph without being interrupted, and hear a long answer back with the full session's context.
Same thread, same agent, same code. Hours later, open OpenClaw at the desk and read everything that was said in the car.
Don't have Clawkie-Talkie yet? Tell your agent to install it.
A voice lane into the OpenClaw session you already have open. Tap the big button to toggle talk mode, land a paragraph at highway speed, and let your main agent reply at length — with the same context the desk already had.
Copy the install instructions and paste them into your
agent. Your agent follows AGENT-INSTALL.md,
inspects the repo first, then installs the persistent
daemon and OpenClaw clawkie-voice-handoff skill.
After install, your agent gives you a host dashboard link. Open the dashboard URL on your phone, bookmark it there or add it to your phone home screen. Choose a Recent OpenClaw Session to open voice; if the list is empty, start or resume one at the desk and refresh.
Press the button to toggle talk mode, speak through the whole prompt without being interrupted. The reply plays back at length while you drive. Hours later, open OpenClaw at the desk and read the canonical transcript of everything that was said.
Push-to-talk you can use to toggle talk mode for a paragraph, full session context on the other end, and a written transcript waiting for you when you sit back down at the keyboard.
Press the big button to toggle talk mode and talk for as long as the thought takes — a paragraph, a few minutes, the whole frame of an essay or an app idea. Nothing cuts you off until you tap again.
Every turn lands in the OpenClaw session you were already working in. No fresh assistant, no lost thread — the agent answers with the code, the prior turns, and the project it already knows.
What you said in the car and what the agent answered are both written into the original OpenClaw/Discord thread. Open it from the laptop later and read everything in order.
The persistent daemon owns the WebRTC link, the STT and TTS calls, and the OpenClaw shell-out. The phone is microphone and speaker; the trusted side stays at home.
Talk through a whole opening, an argument, a piece of training material. Land the idea on the highway, read the transcript at the desk, polish the words you actually said.
Pitch your main agent a new product, an app concept, a feature you've been turning over. Use toggle talk mode for the full framing — let the session you already have push back with real context.
You were mid-feature when you got in the car. Open the dashboard, choose that OpenClaw thread, steer the next decision in long form, and pick up the canonical transcript when you stop.
The whole product runs on your agent's side — the
persistent daemon, the WebRTC rendezvous, OpenClaw infer STT,
OpenClaw infer TTS calls, and the OpenClaw clawkie-voice-handoff
skill that can give you the dashboard URL again later.
Tell your agent to follow
https://github.com/davidguttman/clawkie-talkie/blob/v1.0.0/AGENT-INSTALL.md:
inspect the source first, stop and ask if anything looks
suspicious, harmful, or uncertain, verify daemon/service
status and skill configuration, then report back in plain English.