Everything walkai-talkai does.
The complete feature list of the macOS app, grouped by what happens on the radio, on your Mac and in your AI agent.
Receiving your voice
You press PTT on your handheld; the station radio next to the Mac receives you.
Whisper speech recognition
OpenAI Whisper (whisper.cpp on Metal) turns your voice into text on the Mac, in Turkish or English. The base model is bundled; Large v3 Turbo, the most accurate for Turkish with English technical terms, can be downloaded in the app; on first launch the app asks once before downloading it (about 1.5 GB).
Silence detection
The command is sent automatically about 1.3 seconds after you stop talking. No button on the Mac.
Any audio input
Choose the input device that carries your radio audio (USB sound card, audio interface, built-in input).
Pause and resume listening
Start or stop listening from the station window or the menu bar.
Recorded commands
Every received command is saved with its audio and can be played back in Command History.
Noise guard
Low-frequency rumble, desk thumps and flat noise are filtered out before recognition, so they are never mistaken for a command.
Self-healing audio
If macOS stops the audio engine (device change, sleep and wake, Bluetooth, USB replug), listening rebuilds itself within seconds. No restart needed.
Target Window mode
Your words are typed into the AI agent you already use.
Any target window
Pick the window to type into from the menu bar (Cursor, Claude Code, Antigravity, Terminal…), and reset it anytime.
Automatic typing + Enter
The command is typed with macOS Accessibility and submitted with Enter.
[radio command] tag
Commands that came from the radio are prefixed so your agent knows to answer by voice as well.
Agent starter prompt
Ready-made radio protocol rules (Turkish and English) that you can edit, save, reset and send to the agent in one click.
Delivery beep
A short two-tone beep goes out over the radio when a command is sent: after typing in Target Window mode, and the moment the request goes out in API Mode. You can turn it off in Settings.
Paste test
Simulate a received command to check that typing into the target window works.
API Mode
No IDE needed: your voice goes straight to an AI model and the answer comes back over the air.
Providers
OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Ollama (local) or any custom OpenAI-compatible endpoint.
Profiles
Save several profiles with name, provider, API key, endpoint, model and token limit; choose the active one.
Live model list
Refresh the available models of a provider from the app.
Conversation sessions
Start new sessions, switch between them, keep or drop context and clear messages. A fresh session starts automatically when you change language.
Starter prompt per language
An editable system prompt for Turkish and English, tuned for spoken radio answers.
Prompt test
Send a test question from the app and hear the answer on the radio.
Speaking back over the radio
Answers are synthesized on the Mac and sent into the station radio, which transmits them via VOX.
Neural voice synthesis
English with Kokoro and Turkish with EMA Lightning, running on CoreML and the Apple Neural Engine. macOS voices as fallback.
Spoken language follows your selection
Announcements and speech recognition use the language you selected (Turkish or English). A request to /api/reply can override it for a single announcement.
Radio-tuned audio
Speech is band-limited to the radio band, levelled to a steady loudness and peak-limited without distortion, so it stays clear on FM handhelds.
VOX wake and hold
Adjustable lead-in (0–2500 ms) to open your radio’s VOX, and a pre-speech hold (0–1500 ms).
Advanced audio settings
Output channel (right by default), announcement output level, noise gate and VOX hold-tone level, to match different radios and cables.
Pre-speech sounds
Silent VOX wake, tactical 3-tone, Motorola talk-permit chirp, single beep, double beep, NASA Quindar, soft chime or deep low tone — with preview.
Roger beep and sign-off
Every announcement ends with a roger beep; “over” / “tamam” is added if missing.
Input level balancing
Overly loud radio audio is levelled automatically before transcription; if it keeps clipping, the Mac input volume is lowered step by step.
Never overlaps
Announcements are queued and played one after another.
Device tests
Test buttons on the microphone and speaker cards: transmit a sample announcement, or record from the microphone and play it back. The recording is never sent to the AI.
Local API for agents
Runs on 127.0.0.1:1988 and is reachable only from your Mac.
$
{"status":"transmitting"}
Speak over the radio
POST /api/reply with a short text (and optional language) — any agent that can run curl can talk back.
Station status
GET /api/status: listening state, mode, language, target app, devices and last command.
Control the station
Switch language (/api/language) and operation mode (/api/mode).
API Mode endpoints
Manage AI profiles, fetch models, create and switch conversation sessions, update the starter prompt and send queries.
Station app
A native menu bar app with a station window.
Menu bar control
Listening, operation mode, target window, language, microphone and speaker — all from the menu bar.
Station window
Station, Audio Settings, Target Window, API Settings, Command History, Settings and About tabs.
Turkish and English
Interface and announcement language; optional spoken notice when you switch mode or language.
Command History
The last 100 incoming commands and outgoing announcements with time stamps, playback and a reload button that transmits any announcement over the radio again.
Permission check
Shows whether macOS Accessibility permission is granted.
License and updates
One payment, no subscription.
Lifetime license for 1 Mac
Remove the license from one Mac and activate it on another whenever you change computers.
Lost your key?
Already bought? Enter your purchase email in the app and the license key is sent to it again. Requests are rate limited.
Offline use
Speech recognition and neural voice run on your Mac. The license is checked online from time to time; the app keeps working offline for up to 14 days between checks.
Automatic updates
Signed, notarized updates delivered in the app; small delta downloads.
Use with care
Announcements and the AI’s spoken replies can be heard by anyone listening on the same frequency, and someone else on the channel can also speak commands to your agent (command injection). Make your agent ask for confirmation before irreversible actions. Radio frequencies, transmit power and licensing rules differ from country to country; you are solely responsible for complying with them.
What you need
- A Mac with Apple Silicon (M1 or newer), macOS 13 Ventura or later
- Two handheld radios with VOX (one stays at the Mac)
- An audio cable (2-pin K-plug or 3.5 mm) to your Mac or a USB audio adapter