docs

voice and fn

talk to bise instead of typing. in bise desktop, hold fn from any app. in bise terminal, ctrl+r dictates, and ctrl+r twice starts voice mode.

bise desktop

fn, from any app

keywhat it does
hold fntalk to bise, with what is in front of you; your words go when you let go
fn fnhands-free: talk without holding; fn again stops
fn spacethe spotlight: find an agent, or ask bise by typing

bise's answer shows in a small panel over the app you're in; a long one offers the rest in the window. with bise's window in front, fn fn starts voice mode with the agent in view.

when you hold fn, bise adds what is in front of you to your question: a screenshot of the window in front (never the whole screen), the tab's URL and title (Chrome, Safari, Arc), the open file and your selection (VS Code, Cursor, Xcode, Zed), and the text of the window. settings › context and privacy turns each one off, says how long images are kept, and lists the apps fn never reads.

fn needs three macOS permissions, asked the first time you hold it: see permissions for fn. settings › voice and fn turns hold and fn fn on or off, and sets the language bise hears.

in the composer, ctrl+r dictates, as below.

dictation

ctrl+r in the composer records; any key stops and keeps the text in the composer, ready to edit and send. esc or ctrl+c stops and drops it.

dictation and fn need a speech-to-text model, the voice role. with a Mistral or an OpenAI key, bise picks one for you. settings › voice and fn in bise desktop, /voice in bise terminal, show what is set and change it.

providermodelskey
Mistralmistral/voxtral-mini-transcribe-3, mistral/voxtral-small-transcribe-3MISTRAL_API_KEY
OpenAIopenai/gpt-transcribe, openai/gpt-4o-mini-transcribe, openai/gpt-4o-transcribe, openai/whisper-1OPENAI_API_KEY
Groqgroq/whisper-large-v3-turbo, groq/whisper-large-v3GROQ_API_KEY
ElevenLabselevenlabs/scribe_v2, elevenlabs/scribe_v1ELEVENLABS_API_KEY
Deepgramdeepgram/nova-3DEEPGRAM_API_KEY

in a terminal, bise models voice lists them with the state of each key.

voice mode

voice mode is a spoken conversation with the agent in view: you talk, and it answers out loud.

bise desktop

with bise's window in front, fn fn starts voice mode with the agent in view, or ends it. with no thread in view, bise's own thread opens and voice mode starts there.

bise terminal

ctrl+r twice starts voice mode.

keyin voice mode
spacesend now
hold spacekeep the floor, or cut in when it talks on speakers
mmute
tabtype instead
escleave voice mode (ctrl+c too)

with headphones it listens hands-free. on speakers you hold space while you talk, so it doesn't hear itself. today only Mistral's voice model speaks, so voice mode needs a Mistral key.

by default it reads aloud what needs you and what you asked. the first time, voice mode says who hears you: your audio goes to the speech-to-text provider, and the text it speaks goes to Mistral.

settings

/voice in bise terminal changes all of these; bise desktop's settings › voice and fn has the language. they live in [voice]:

~/.bise/config.toml
[roles]
voice = "mistral/voxtral-mini-transcribe-3"   # dictation's model

[voice]
language = "fr"                          # unset: detected
vocabulary = ["bise", "config.toml"]     # words to spell right
listen = "auto"                          # auto (hands-free with headphones, hold on speakers), hands-free, hold
tts_voice = "…"                          # the voice it speaks with; unset: the default voice
speed = 1.0                              # 0.8 to 1.6
read_aloud = "needs"                     # needs (what needs you + what you asked), all, nothing
sounds = true

BISE_VOICE_MODEL sets the dictation model for one session.