I built Clawkie Talkie for the thoughts that are too long to type and too useful to leave in my head. The canonical use case is driving: I have a feature, essay, app idea, or training outline already alive in an OpenClaw thread, then I get in the car and want to keep pushing it forward without pretending a one-line voice command is enough.

Clawkie Talkie is not a separate assistant. It is a voice lane into the same OpenClaw session: same thread, same context, same agent, same written transcript waiting for me when I get back to the laptop. I say "switch to voice," OpenClaw gives me a handoff link, I open it on my phone, toggle talk mode, and talk through the whole thought without being interrupted.

How it works:

The phone/browser is the microphone, transcript display, push-to-talk control, and reply speaker.
A local Clawkie Talkie daemon on my machine holds the trusted side of the connection and talks to the local openclaw CLI.
The handoff URL carries routing for the current OpenClaw session, so the spoken prompt lands back in the original thread and the agent reply is played back over the voice link.
The web app uses WebRTC for the voice session; the hosted service helps the browser and daemon rendezvous, but OpenClaw credentials and provider keys stay on the local machine.

Why the daemon matters:

The browser UI is intentionally lightweight. It does not get my OpenClaw credentials, model provider keys, or long-lived local configuration. The daemon is the trusted boundary: it handles STT, the OpenClaw agent turn, TTS, and transcript/reply delivery through the same local OpenClaw setup I already use at the desk.

Installability:

The install path is built for agents to run. The public instructions have an agent install flow that fetches the source, sets up the persistent daemon, installs the OpenClaw handoff skill, verifies STT and TTS, and checks that the daemon and skill agree on the same host ID. The goal is that a nontechnical user can ask their agent to install Clawkie Talkie, approve only the genuinely human steps, and end up with "switch to voice" working from any existing OpenClaw session.

Clawkie Talkie exists because some of the best prompts are not prompts at all. They are ten uninterrupted minutes of thinking out loud, with the full project context already loaded and waiting.

david.app // project terminal

$ ls ./projects