Voice command for your coding agent
A realtime voice AI that masters your coding agent, not just another chat box. It commands Claude Code from the bridge today, with Codex coming next: by voice, remotely, every project under way at once.
A live run from the bridge: speak the order, watch Claude Code work the deck in real time. Ninety seconds, sound on.
The act, station, or duty of directing the steering of a ship. To have the conn is to hold command of a vessel's course, directing the helm from the bridge.
Conn turns your voice into command of your coding agent. You give the orders; a crew of AI sessions carries them out across every project, and you see the whole deck from one place, from anywhere.
Conn runs on iPhone and on the desktop, and both are live now. The wide desktop page is the full deck on your Mac; the iPhone app is the same bridge in your pocket. Speak the orders, watch the work stream in, stop a run mid-flight, from whichever screen is in your hand.
Conn is a realtime voice AI that masters your coding agent. It runs on OpenAI's realtime model, purpose-instructed to actually fly the agent: it knows the tools, the permission tiers, the session model, when to push work and when to ask. You speak in plain language; it translates that into real agent operation and reads the results back. Conn drives Claude Code today, with Codex support on the way.
Claude Code now · Codex coming nextYour Mac runs the agents; your phone runs you. Kick off real work from the street, the trail, the queue. On iOS, Conn holds the audio session alive in the background: lock the screen and keep talking, the conversation stays live with the phone asleep in your pocket.
The iPhone app and the desktop page hold a single bridge between them. Pick a session on your phone and it is already there on the desktop; start one on the desktop and it surfaces on the phone. The watch hands off clean, live voice and all. Take the call at your Mac, walk out the door still talking, finish it on the trail. Conn never loses the conn.
Run up to eight agent sessions at once, one per project, side by side, working in parallel. Each shows as a station you can switch to with a word; the active one has the conn, the rest keep running. No more single-file, one thing at a time.
Save a session and it is stowed whole: the full conversation and the complete work log. Come back hours or days later, retrieve it, and Conn picks up exactly where you left off, context intact. Every watch is recorded; none of it slips overboard.
Cut in mid-task to redirect, take over, or stop a run cold, by voice or a tap.
Speak when your hands are full; type when a command needs to be exact.
iOS keeps the audio session alive. Talk with the screen dark and the phone away.
Risky actions surface for approval. You stay in command of what the agents touch.
Every tool call, every result, streamed to the deck as it happens.
Conn drives the coding agent already installed on your own Mac. Your code never leaves it.
Conn is bring-your-own-key. You pay OpenAI, xAI, and Anthropic directly for what you use; Conn never bills for usage. Set your watch below and estimate the monthly cost of commanding your agents by voice.
Flat, every month. No usage fees, no markup, no per-seat charge. That is the whole bill from us, and it does not move.
Estimates only. The Voice and Agent figures below are not quotes. They are estimates built from publicly available provider pricing and typical usage assumptions, and your real cost will vary with how you actually work. This is what you pay OpenAI, xAI, and Anthropic directly, through your own API keys and subscriptions. It is your own AI usage, it is not paid to Conn, Conn does not mark it up, and you would pay it to the providers no matter how you reached the agent.
Estimate only. Conn is a flat $10 / month. You pay OpenAI, xAI, and Anthropic directly for AI usage; Conn does not mark up provider usage.
Subscription plans are a flat monthly fee. The agent line does not scale with hours; only the voice line does. Anthropic describes Max as 5x and 20x more usage than Pro; OpenAI describes Codex limits as message counts in a 5-hour window. Conn does not estimate an included hours or token figure for any plan.
Codex subscription promo multipliers run through May 31, 2026; base plan limits apply after. AI pricing changes often. Verified May 2026.
One commander. A crew of agents. Every project under way, through the coldest night.