Talk to your coding agents, and see what they mean.
Command Bridge gives your agents a voice and a canvas. You speak; they answer out loud and draw the answer while they explain it. It runs on your own machine, with no account and no speech API.
curl -fsSL https://raw.githubusercontent.com/JuanjoFuchs/command-bridge/main/install.sh | bash
Then tell your agent: run command-bridge describe and follow it. Windows and PyPI installs are in the README.
Why I built it
I kept hitting the same wall with my agents: getting what I meant in was slow, and getting what they meant out was slower.
Typing to an agent
I type a request, read a wall of text back, and type again. Most of my day with AI agents went into that loop.
Talking is three times faster
Dictation let me explain my reasoning instead of abbreviating it, and the results got closer to what I imagined.
Reading is the bottleneck
Reading the answer back is the slow half. What if I could hear it, and see what it means? Intent and understanding per minute
So I built Command Bridge
Each agent keeps working in its own terminal. Command Bridge carries my voice to it, and its voice and drawings back to my laptop or my phone. The agent points at the part it is talking about. The agent points
One voice, a whole team
I talk to one agent, my chief of staff. The others work under it and brief it, so a dozen agents never talk over each other. I let my AI agent manage my attention
A voice turn is measured end to end, so I can see where the waiting goes. From 15 seconds to under 1
Local
Speech recognition, the voice and the voiceprint run on your machine. No account and no speech API. A CPU is enough, and a GPU makes it faster.
Any agent
The tool holds no model. If your agent can run a command, it can talk to you. Claude Code, Codex and Grok have each driven it unchanged.
One command to learn it
command-bridge describe hands your agent the whole contract, so it sets itself up and knows how to listen.
Writing
Command Bridge grew in public. These posts tell how each piece came about.
How I watch what my coding agents are doingA sunflower view of which files my agents read and write, which commands they run, and who is idle.
How I cut my local voice agent's wait from 15 seconds to under 1Speech on the GPU, and transcribing while I talk.
I let my AI agent manage my attentionI talk to one agent now, and my team went from six to 16.
Deixis: the agent pointsPointing at the exact element while it speaks, across several agents.
Bandwidth: intent and understanding per minuteTwo separate limits, and the four steps that widened both.
Launching voice-tunnelWhere it started: a two-way voice channel from any coding agent to a phone browser.
Get it
Open source under the MIT license. One line installs it, then your agent sets up the rest.
curl -fsSL https://raw.githubusercontent.com/JuanjoFuchs/command-bridge/main/install.sh | bash