Skip to main content

Audio and video in chat

Alongside image understanding, visitors can attach an audio or video file in the chat and ask about it. The chat composer's + (add) menu offers:

  • Upload image — ask about a picture (see Image understanding)
  • Upload audio — ask about a sound or voice recording
  • Upload video — ask about a video clip
  • Attach file — add a document to the conversation

Audio and video pickers are mutually exclusive with each other on a single message — attach one media file per turn.

Uploading audio​

Choose Upload audio from the + menu, pick a file, and ask your question.

FormatsMP3, WAV
Maximum size25 MB

On the turn the audio is attached, the chatbot answers from the audio itself — the knowledge base is not searched for that message. There are two ways the audio is understood, depending on the chatbot's setup:

  • A model that listens to the audio directly. If the chatbot uses Bring Your Own Inference (BYOI) and its chat endpoint is audio-capable, the audio file is sent to that model, which listens to it and answers in one step.
  • Transcribe, then answer. Otherwise, the audio is transcribed to text and the chatbot answers from the transcript. Transcription uses your speech-to-text endpoint when one is assigned, and WizChat's transcription service otherwise.

Either way, a transcript of the audio is captured and carried into the conversation, so your follow-up questions run through the normal knowledge-base pipeline with the audio's content available as context.

Data residency

On chatbots configured for zero data retention or EU residency, audio requires self-hosted (BYOI) speech-to-text and text endpoints. If those aren't configured, audio is unavailable on that chatbot and the message returns an error rather than falling back to the platform service.

Uploading video​

Choose Upload video from the + menu, pick a file, and ask your question.

FormatsMP4, WebM, MOV
Maximum size100 MB

Video understanding is BYOI-only: it works only when the chatbot's Bring Your Own Inference chat endpoint is a self-hosted, video-capable model. The uploaded video is made available to that endpoint, which watches it and answers your question; the knowledge base is not searched for that message.

If the chatbot has no video-capable endpoint, the message returns:

Video understanding requires a self-hosted video-capable endpoint on this chatbot.

Why BYOI-only?

WizChat's platform models do not accept video input, so there is no platform fallback for video — a self-hosted video-capable endpoint is required. Audio, images, and text are unaffected.

Where files are stored​

Uploaded audio and video files are stored in a private, per-user location and are never made public. The model reads a file through a short-lived signed URL (valid for about 15 minutes) that only grants access to that one file. For BYOI video, the signed URL is handed to your own endpoint, which fetches the video itself — WizChat streams the URL, not the video bytes.

See Data retention for how uploaded media is retained and deleted.