Dictate a message or talk with Voice

3 minute read · Chats & Local Agents

Use dictation to write into a chat, or Voice conversation to hear spoken responses. These are separate controls in Settings → Voice. Choose the microphone, speech model, language, conversation voice and start/stop phrases there.

• Dictate into the message field
• Connect cloud speech
• Start and stop a conversation
• Check a short example

Dictate into the message field#

1. Turn on Enable voice dictation in Settings → Voice.
2. Choose a Speech model. Local models run on this computer after their download and preparation finish. The picker shows download size, readiness and memory information. Cloud models use your own OpenAI API key.
3. Choose your Microphone and Language. Device names can appear only after microphone access is allowed. Choose Български for Bulgarian mixed with English technical terms.
4. Return to the chat and press its microphone button or the configured dictation shortcut. In Toggle mode a press starts or pauses capture; in Hold mode the shortcut records while held and keeps the text when released.
5. Review the text. Press Esc to pause, edit the draft and send it normally. The Vortex send end phrase is available for spoken submission. Pausing or closing input keeps the visible text; silence alone is not a request to send it.

Where the selected local model supports translation, Write in → English translates speech into English on this computer. Other models may expose only the original language. Clear selection unloads the local model and requires choosing a model again before dictating. A cancelled download or failed preparation is not a ready model; use its displayed Retry action.

Connect cloud speech#

Use Add API key in Settings → Voice for cloud dictation or Voice conversation. The key is stored encrypted on this computer for the current account. Replace key and Remove key manage it there; never put the key in a chat message.

Cloud speech sends audio to OpenAI and uses the API account's access and billing. Connecting a coding-agent account does not configure this separate Voice key. Local dictation does not need a cloud speech key once its model is installed. Local speech recognition does not make a message sent to an online coding agent stay offline.

Start and stop a conversation#

Turn on Enable Voice conversation; it starts off in a new profile. Select a conversation Model and Voice, then use the Voice control to begin. The built-in synthetic voice choices include Alloy, Ash, Ballad, Coral, Echo, Sage, Shimmer, Verse, Marin and Cedar. Model and voice changes apply to the next session.

Listen for the start phrase is a separate opt-in and is off by default. It recognizes phrases on this computer; dictation pauses wake listening. The defaults are Start voice mode and Stop voice mode. Enter distinct custom phrases in Settings → Voice, or leave the fields blank to use the defaults. Say the stop phrase on its own to interrupt playback and turn off the microphone. Stopping also disables continued wake listening; enable it again deliberately when needed.

Watch the listening, connecting, speaking or error feedback during use. An enabled setting alone does not prove the microphone started. Turning Voice conversation off stops the conversation and wake listening. Stopping speech does not undo changes an agent has already made; review the chat before retrying work.

Check a short example#

Dictate a harmless sentence, pause and confirm it remains editable without being sent. For conversation, start a short response, use Stop while it speaks and verify the sound and microphone activity stop. If input fails, check the selected device, model readiness, API-key status and Settings → System permissions. Preserve the draft and read the actual error before retrying.

Related guides: settings, agent permissions, models and usage.