Your agents, out loud.
Brainstorm, write and build with full-duplex voice.
Never a spectator — see every step, weigh in on every decision, cut in and steer anytime.
Can't reach GitHub? Use a mirror
Live OnBoard is an open-source AI co-worker for macOS.
It works natively across your Mac, powered by the agents you already run — Codex, Claude Code — on your subscription or API key.
Full-duplex voice
It's never too busy to talk.
Two models run at the same time: one holds the conversation, the other does the work. The one you're talking to is never the one that's busy.
So nothing puts you on hold. Interrupt it, change your mind, or ask how far along it is — at any point, mid-sentence — and the work turns with you.
The agents underneath lose none of their reach.
How full-duplex worksYour ideas, amplified
Half an idea is enough.
You don't have to arrive with a finished thought. It asks about the parts you left open, you answer in a sentence, and what comes back is clearer than what you brought it.
You're both working from the same picture the whole way through — so what ends up on the page is what you meant, not its best guess at it.
System-wide
It reaches the rest of your Mac.
110 tools20 different capabilities
See every capability
Yours to assemble
Free to download. Free to run.
Speech in, speech out and the model that keeps the conversation going all run on your Mac, with nothing metered by usage.
The actual work goes to the agents you already subscribe to. Claude Code, Codex — whichever you have, your existing plan covers it.
Every layer can be replaced on its own: change the local model, switch the agent behind it, or point any one of them somewhere else without disturbing the rest.
See what runs whereSpeech in and out
- On your Mac
The model that talks
- Qwen3.5-4B
- Qwen3.5-2B
- Gemma 4 E4B
- Phi-4-mini
- Qwen3.5-9B
- gpt-oss-20b
- Gemma 4 26B
- Qwen3.5-27B
- Gemma 4 31B
- gpt-oss-120b
The agent that works
- Codex
- Claude Code
- OpenAI-compatible API
- Anthropic-compatible API
- Your own agent
Getting started
One file. It sets itself up.
Open the .dmg and that's the last decision you have to make. It fetches the models, the speech engines and everything else it needs, and shows you where it's up to the whole way.
No terminal, no package manager, nothing to download by hand.
- Live OnBoard
- Speech engines
- Qwen3.5-4B
- Dependencies
And more
A few more, at launch.
A few of the things Live OnBoard does aren't on this page — they're better seen than read about.
All of it will be shown at launch, along with whatever else is ready by then.
There’s a lot more to build.
Live OnBoard is early, and so far it's one person's work — which means rough edges here and there, and plenty I'd do better with more hands on it.
Project CommonSwell is where that work happens in the open. If any of this looks like something you'd want to build, I'd be glad of the company — twenty years of shipping software or still at school, it's the same door.
About the project