Sidekick

Talk mode

Hands-free conversation, interrupting, stopping, and the band.

/sidekick talk toggles talk mode. While it is on, one listener runs the whole time, including while the sidekick speaks and while it works.

The loop

StepWhat happens
ListenThe microphone opens. Partial words show in the band.
SendAfter a 1.4 second pause, what you said is sent as your prompt, verbatim.
WorkClaude works as usual. The band says working….
SpeakThe reply's first one to three sentences are read aloud in the sidekick's voice.
ChimeOne soft note: your turn. Listening already resumed.

Interrupting

Talk over the sidekick. The first two words that are not its own cut the speech and your sentence becomes the next prompt. Its own voice reaching the microphone is recognized too, so the mod filters words that match what it just said; see How it works.

Voice commands

SayEffect
"stop", "wait", "hold on", "never mind"Cancels the running turn (nothing is sent). Also cuts speech.
"end talk", "stop talking", "that's all", "goodbye"Leaves talk mode. Lower chime.

Everything else is a prompt. Speaking while the sidekick works queues the sentence as the next prompt.

The band

A row above the prompt shows the state and the keys:

🔊 Rudy is speaking… talk over to interrupt   s skip   m mute   x end talk
🎙 listening…  what about the docs            m mute   x end talk
🦊 Rudy is working… say "stop" to cancel
❓ Deploy where?  1) Staging  2) Production

s skips the current speech, m toggles speech, x ends talk mode, l listens now when idle.

When it stops on its own

Three listening rounds in a row with nothing heard end talk mode, with a toast. /sidekick talk resumes. A new session always starts with talk mode off.

Tuning

The pause that ends an utterance is --silence 1.4 and the longest utterance --max 60, both in the listener's arguments in hooks/register.tsx. The interrupt threshold is the two-word rule in the same file. Add --lang en-GB style codes to pin a recognition language; the default is your macOS locale.

On this page