Research, voice & media
Dictation, live voice and Voice Assistant
Dictate text, speak live with Codex or use IXI through voice and an optional wake phrase.
App version 0.7.934 · Updated
Get started
- Choose the appropriate route: dictation for text input, Live Voice in a Codex chat or Voice Assistant for IXI.
- Check microphone, output and the connection required by that route in its settings.
- Start deliberately, check the visible connection state and end the conversation with its Stop action.
Dictation and text-to-speech
Dictation turns spoken words into text input. Voice settings expose the configured transcription service, such as Whisper or browser recognition. Review the transcription before sending it as a task.
Text-to-speech has separate settings for supported voices, language and speed. Services such as ElevenLabs require their own access and may incur usage charges. Dictation does not activate Claude and Codex collaboration.
Codex Live Voice in a session
Open Live Voice in a Codex chat. Its panel offers voice selection, extra instructions, audio devices, microphone mute, output testing and a Stop action. Availability depends on your Codex connection and installed CLI version.
Speech and responses belong to the Codex session. Text input and approval cards remain accessible while the panel is open. One live voice connection runs at a time; the panel reports when another session is using it. Ending the conversation, closing the panel or leaving the chat view stops voice and releases local audio devices. The Codex text process remains available.
Talk to IXI and its tools
Voice Assistant connects speech to IXI's agent chat, tools and memory. The GPT-Live voice connection requires a separate OpenAI API key and is billed separately from a ChatGPT subscription. Requests involving tools, files, projects or memory are delegated to the normal IXI agent and additionally require its configured ChatGPT/Codex access. The voice key is encrypted on this device and is not included in account synchronization.
Choose a voice and audio devices, test the connection and set an inactivity timeout. Spoken messages and responses appear in agent chat. Usage displays recorded voice time, cost information and separately tracked agent usage. Delegated work uses the limits of IXI's connection; external tool services may have their own charges.
Wake phrase and other voice routes
Optional local wake recognition on Windows requires the German speech recognizer and uses the Windows default microphone while IXIOM is running. Configure the phrase and sensitivity. Local recognition by itself does not keep a paid voice connection open.
Grok has a separate voice interface using its provider connection. Collab voice is for conversations with people and supports push-to-talk or voice activation. Choose these routes according to the conversation and task.