Voice mode
On your Mac, Pong can read each Ping aloud and take your answer by voice, so you can answer with your hands full or your eyes elsewhere.
Turning it on
In Settings, under Voice, turn on Read Pings aloud and answer by speaking. The first time, macOS asks to let Pong use the microphone and speech recognition.
With Listen when a Ping appears on, Pong reads each new Ping and then listens. With it off, press the microphone on the card when you want it.
What you hear
Pong reads the question, then the answers, numbered, and says which one the agent recommends. While it listens, a level meter on the card moves with your voice.
What you can say
- An answer by its number: “the first one”, “two”, “two and three”.
- “Recommended” for the agent's pick, or “yes” and “no” when there are two answers.
- “Explain” or “what is this about?” for a short explanation of the question.
- “Repeat” to hear the Ping again.
- “Stop” to stop listening.
With a model chosen, you can also just talk: “let's do just the users table, and add a test for it” picks the right answer and writes the rest as your text.
Before it sends
Pong never sends what it heard straight away. The answer it understood lights up on the card, any text is written into the reply box, and a short bar fills over about three seconds. Then the Pong is sent.
Say “wait” to pause it and “go” to carry on, or “cancel” to clear it. Clicking the card stops the bar and leaves the answer for you to finish by hand.
Listening ends when a Pong is sent, after 30 seconds of quiet, or when you say stop.
Models
Without a model, Pong understands numbers, “recommended”, “yes” and “no”. A model lets it understand ordinary speech. Choose one in Settings under Voice:
- Hosted: Pong's own model, which comes with Pong Cloud.
- Endpoint: any OpenAI-compatible server, including a model running on your own computer.
- Anthropic API key: your own key, kept in the macOS Keychain.
Test sends a sample to the model and shows what it understood.
What leaves your Mac
Your voice is turned into text on your Mac, and audio never leaves it. With a model, only the words you said go to it, with the Ping's question, answers and context. The hosted model reads them for that one request and keeps nothing. See Privacy and security.