# TWIN AI desktop guide ## What is TWIN AI? TWIN AI is an open-source personal AI twin for Windows. Its desktop studio uses your CV, portfolio and project notes to answer questions in writing, in a voice based on your recordings, or through a synchronized virtual camera. [Build a profile from your work](https://twinaistudio.com/features/profiles/) ## Which parts run on my PC? Speech transcription, voice synthesis and avatar rendering run locally with the installed models. The app sends the question and relevant profile text to NVIDIA for the written reply, so generating answers requires an internet connection and an NVIDIA API key. [Check Windows and hardware requirements](https://twinaistudio.com/download/) ## Can I start without recording my voice or face? Yes. Choose text-only replies and use selected text as the question source. Highlight a question in a supported Windows app to get a written reply using your profile. This setup skips transcription, voice synthesis and the camera. [Set up selected-text replies](https://twinaistudio.com/features/selected-text/) ## How do voice and video reach a call? The Windows app can listen to questions on your headset and send spoken replies through VB-CABLE. For an avatar session, select Unity Video Capture as the camera in the receiving app. Choose MuseTalk or the experimental W2L engine before loading the twin. [Connect your headset and virtual camera](https://twinaistudio.com/features/devices/) ## Setup Start with the Windows release and add the components your chosen output needs. ### Connect the reply service Set NVIDIA_API_KEY in the local environment, then restart the app. Questions and relevant profile text go to NVIDIA for replies. ### Choose local models Install transcription for spoken questions and the voice model for speech. For video, choose MuseTalk, experimental W2L, or install both and select one before preparing a session. ### Finish driver setup Use the vendor installers for the virtual camera and audio cable. Restart Windows if prompted, then reopen the studio and your call app. [Illustrated Setup guide](https://twinaistudio.com/features/) ## Profiles Each profile keeps its own experience, project notes, languages, voice collections, and avatar collections. ### Write the background Create a profile, add your role and CV details, and save. Add knowledge cards with the decisions, trade-offs, and outcomes behind your projects. ### Choose references and language Select a voice and an avatar collection. The supported conversation language sets recognition and reply language together. ### Manage collections Create a separate voice or avatar to try new recordings. Preview its prepared reference, or use Reset recordings on the selected collection when you intend to start it over. ### Switch or move a profile Stop active work before switching. Export a profile to transfer it, then import it on the other PC and check devices and prepared media. [Illustrated Profiles guide](https://twinaistudio.com/features/profiles/) ## Voice recordings Open Recordings with the intended profile and voice collection selected. A quiet microphone recording gives the speech model a useful reference. ### Record the owner reference Select the microphone and recording language. Keep the mic in one position and read the displayed passage. ### Listen before accepting Play the take back. Retake clipped words or noisy endings. The English guided set has 50 prompts and allows up to three accepted takes per prompt. ### Prepare and preview Prepare the selected voice from accepted material. Preview a familiar sentence to check pronunciation and tone. [Illustrated Voice recordings guide](https://twinaistudio.com/features/voice-recordings/) ## Avatar recordings The avatar guide covers ten takes, from a neutral reference to speech, head movement, and facial expressions. ### Frame the camera Choose Avatar recordings and open Preview camera. Frame your head and shoulders, keep the lighting steady, and place the prompt close to the lens. ### Follow the cue Record the requested action. Start and finish in a neutral pose so the idle and speaking footage can join smoothly. ### Review and prepare Watch the complete take before accepting it. When the required references are ready, prepare the selected avatar collection. [Illustrated Avatar recordings guide](https://twinaistudio.com/features/avatar-recordings/) ## Devices Use the Devices workspace to identify the physical inputs and virtual outputs before starting a live session. ### Pick the question source Use your microphone for a direct conversation or your headset playback device to hear questions from a call. Headset capture includes other sound playing on that device. ### Route speech The studio sends replies to CABLE Input. Choose CABLE Output as the microphone in the receiving app. Enable headset monitoring if you also want to hear the reply. ### Check a device Refresh or search the device list if a saved device is unavailable. Open Audio diagnostics to test the chosen input or monitor output before a session. ### Route video Choose Unity Video Capture in the receiving app. Prepare an avatar-enabled session in TWIN AI to supply idle and speaking frames. [Illustrated Devices guide](https://twinaistudio.com/features/devices/) ## Conversation Preparation loads the selected models. Listening starts when you explicitly start the input. ### Choose the output Pick text only, voice only, or voice with avatar. Select the avatar engine before Prepare twin. Stop and unload to change output mode. ### Ask and listen Type a question, speak into your microphone, or capture headset audio. In session settings, adjust the listening pause for longer questions and choose a delivery mood. ### Switch sources during an interview Pause listening, choose your microphone, headset or selected text, then resume. The twin keeps the loaded models and conversation history. A partial question is discarded and an answer already in progress can finish. ### Control the session Choose Until I stop or a timed limit. Check the current stage and turn timings. Stop & unload releases the active models and camera. [Illustrated Conversation guide](https://twinaistudio.com/features/conversation/) ## Text and selection Selected text bypasses audio capture. Pair it with text-only replies for a session that needs neither a voice model nor a camera. ### Prepare text mode Choose Text-only replies and Selected text before preparation to skip the transcription model too. ### Select the question Start selected text, then highlight up to 1,500 characters in a supported Windows app. Release the selection to submit it. ### Combine with listening With microphone or headset input, enable Also answer selected text. Speech takes priority, and selections made during speech or an answer are discarded. [Illustrated Text and selection guide](https://twinaistudio.com/features/selected-text/) ## Read a script Use a repeatable sentence to check timing, voice quality, or routing before a call. ### Choose a media output Prepare a voice or avatar session and open Read a script in Conversation. ### Enter the line Write the words you want spoken and choose the countdown. The studio prepares and checks delivery before counting down. ### Check delivery Listen in the headset and check the cable input meter in the receiving app. For avatar output, check the virtual camera at the same time. [Illustrated Read a script guide](https://twinaistudio.com/features/script/) ## Control widget The desktop widget brings the current question, generated reply, and session controls together above your other windows. ### Read along Watch the listening, processing, and playback states. During speech, the widget highlights the estimated current word. ### Review a turn Browse recent questions, copy the displayed reply, and use Back to live to follow the newest answer. ### Use quick controls Adjust text size, pin or compact the widget, prepare or stop the twin. Capture exclusion depends on the Windows capture method. [Illustrated Control widget guide](https://twinaistudio.com/features/widget/) ## Sessions Check recent turns and open the last locally saved session when transcript saving is enabled. ### Review a question Read the recognized question beside its generated reply. Check text response timing or the separate speech and media timings where available. ### Open the saved session Use Open saved session to inspect the text and delivery results. The session log contains no audio or video recording. ### Choose what to keep Set the transcript preference in Settings before preparing the next session. [Illustrated Sessions guide](https://twinaistudio.com/features/sessions/) ## Settings and tray Configure the interface, privacy preferences, service status, and Windows tray behavior from Settings. ### Set the appearance Pick light, dark, system, or custom colors. Choose the accent, font, text size, and interface language. Speech language is selected in Profiles. ### Check privacy and status Choose whether to keep local conversation text and inspect the service status before the next run. ### Set global interview shortcuts Under Interview shortcuts, click a binding and press your preferred keys. The defaults are Ctrl+Alt+1 to load, Ctrl+Alt+2 to start or pause listening, Ctrl+Alt+3 to cycle sources while paused, and Ctrl+Alt+4 to stop and unload. They work while another window has focus. ### Use the Windows tray Reopen the studio, jump to Profiles, or stop listening and unload from the tray. Stop all or Escape stops app-owned activities. [Illustrated Settings and tray guide](https://twinaistudio.com/features/settings/)