walkai-talkai

walkai-talkai is a macOS app that turns any handheld two-way radio into a push-to-talk voice bridge for AI coding agents. You key up on a walkie-talkie, speak a command, and the app transcribes your voice locally on the Mac, types that command into an agent such as Cursor, Claude Code or Antigravity, and then broadcasts the agent's answer back over the radio with a neural voice. It is built for developers, radio users and anyone who wants hands-free control of an AI agent while away from the keyboard. It works with Baofeng, Quansheng, Motorola, Kenwood or any radio that has VOX and an audio jack, running on Apple Silicon Macs with macOS 13 or later. Most work with an AI coding agent happens at a keyboard: you type a prompt, wait for the agent, read the answer on screen and then type again. walkai-talkai changes that loop by making the radio the microphone and the speaker, so the interaction becomes a transmission rather than a chat window. The website presents the product as "built for the field", and it emphasizes that the app types into the exact window you chose even while you are away from the keyboard. Because speech recognition and speech synthesis both run on the Mac, your radio audio never leaves the machine; the internet is used for the license check and, in API Mode, for the AI provider you choose. The app keeps working offline for up to 14 days between license checks, which matters when the station sits somewhere without a reliable connection. On the input side, walkai-talkai treats the radio as the microphone. Audio comes in through a simple cable — a 2-pin K-plug or 3.5 mm connection into the Mac or a USB audio adapter — and OpenAI Whisper running locally on Metal converts speech to text. The app supports Turkish and English, and it shows live waveform monitoring so you can confirm the station is hearing you before you commit to a command. Inside the Station dashboard the live spectrum displays incoming audio in green and your reply in orange, with each message appearing underneath for a few seconds. Because recognition happens on the Mac rather than in the cloud, the audio path stays on the machine from end to end. Once a command has been transcribed, walkai-talkai offers two ways to deliver it. In Target Window mode you pick a target window once — Cursor, Antigravity, Terminal or any app — and every transmission is typed into that window through macOS Accessibility automation, then submitted with Enter and tagged as a [radio command]. Because the app drives the window itself, it keeps working even while you are away from the keyboard. In API Mode the app skips the IDE entirely and calls a model provider directly: OpenAI, Anthropic, Gemini, DeepSeek or Ollama. That makes the radio either a remote control for an existing coding session or a standalone voice front end for a model, without changing hardware. The return path is what makes the loop feel like radio rather than a chat app. Any agent that can run curl can answer by POSTing a short summary to 127.0.0.1:1988/api/reply, and the endpoint is loopback-only, so nothing is exposed to your network. The app then speaks that text using CoreML neural TTS — Kokoro and EMA models running on CoreML and the Apple Neural Engine — for fast, high-quality radio announcements. To open a radio's VOX circuit before the first syllable is spoken, walkai-talkai sends a rising 880 → 1100 → 1400 Hz wake-tone preamble, and it closes the transmission with a roger beep. The result is an answer that arrives as a proper radio transmission rather than a truncated squawk. Seen end to end, the product is a five-step loop: key up, transcribe, dispatch, reply and broadcast. You press PTT and speak; your radio's audio-out feeds the Mac through a cable; Whisper converts your voice to text; the command is typed into the agent and sent; the agent POSTs a short summary back to the local endpoint; and CoreML neural TTS speaks it into the radio's VOX line with wake tones and a roger beep. The Station UI is deliberately calm and focused, with sections for the Station dashboard, Audio I/O, Automation, API settings, Command Logs and Diagnostics, a one-click start/stop for listening, and a switch between Target Window and API mode. Low latency is called out as a core quality of the app, which matters when a reply has to fit between transmissions. The benefits follow directly from that design. You can operate an AI coding agent without sitting at the keyboard, which suits anyone who is moving around a room, a lab or a workshop while a build or a task runs. Because Whisper and the neural voice run locally, the radio audio stays on your Mac, and the loopback-only agent API keeps the reply channel off your network. The app works with radios you may already own rather than requiring dedicated hardware, and it is agent-agnostic: anything you can type into is a valid target, and any agent that can run curl can talk back. Speech recognition and neural voice cover both English and Turkish, and a lifetime license can be moved to a new Mac whenever you want. Concrete setups show how the loop is used. A developer leaves one radio at the Mac and carries the second one, then keys up to ask an agent to look at a problem — the site's example transmission asks the agent to check why a buy button does nothing when clicked. Another developer uses Target Window mode to type a quick instruction into Cursor or Antigravity without returning to the desk, letting the agent work while they are away from the keyboard. In API Mode, the same radio can query OpenAI, Anthropic, Gemini, DeepSeek or Ollama models directly, with no IDE in the path, which is useful when you only want a spoken answer. Radio operators who already run VOX-capable handhelds can reuse their existing Baofeng, Quansheng, Motorola or Kenwood gear, and Turkish-speaking users can work entirely in their own language for both recognition and the spoken reply. Hardware requirements are deliberately modest: any handheld two-way radio with VOX and an external audio port, a standard 2-pin K-plug or 3.5 mm audio cable into the Mac or a USB audio adapter, a second radio so one stays at the Mac and one travels with you, and an Apple Silicon Mac (M1 and newer) on macOS 13 Ventura or later. Pricing is a single lifetime license priced at $19.99 as an early-bird rate against a $29.99 regular price, excluding tax, with any applicable sales tax or VAT added at checkout. Stripe acts as Merchant of Record for every order, issuing the receipt and handling tax, while walkai-talkai provides the software, the license and support. One license activates one Mac, and you can deactivate it and move it to a new Mac whenever you want; free updates are included. In short, walkai-talkai is a narrow, well-defined tool: it turns a walkie-talkie into the microphone and speaker of an AI coding agent on macOS, keeps the audio processing local, and closes the loop with neural voice, VOX wake tones and a roger beep. If you want hands-free, radio-style control of Cursor, Claude Code, Antigravity, a terminal agent or a direct model API, that is exactly the loop it builds.