You can attach files to any chat message so the AI agent can analyse them alongside your question. Attachments are processed before the message is sent and their content is included in the conversation.
How to attach
- Click the paperclip icon in the chat input toolbar and pick files from your device.
- Or drag and drop files directly onto the chat window.
You can attach multiple files at once. Each file shows a progress spinner while it uploads; the message is only sent once every attachment is ready.
Supported file types
| Category | Extensions | What the AI does |
|---|
| Images | PNG, JPEG, GIF, WebP | Looks at visual content — layout, charts, photos, text inside the image. |
| Documents | PDF, DOCX, XLSX, PPTX, TXT, Markdown | Reads the text content, including tables and headings. |
| Code | JS, TS, Python, and most common source formats | Reads the source as plain text with syntax awareness. |
| Audio | MP3, M4A, WAV, OGG, WebM audio | Transcribes the audio track and hands the text to the agent. The raw bytes never reach the chat model. |
| Video | MP4, MOV, MKV, WebM, AVI, MPEG, 3GP, M4V | Extracts the audio track, transcribes it, and hands the text to the agent. Visual content is not sent. |
Audio and video transcription
When you attach an audio or video file, the platform runs a server-side transcription pipeline before the message is sent:
- The file is compressed to Opus (and chunked if large) so it fits the transcription model’s input limit.
- Each chunk is sent to the organisation’s configured transcription provider model (e.g. OpenAI Whisper or a self-hosted Whisper-compatible server such as faster-whisper-server, vLLM, or LocalAI).
- The returned transcript is attached to the message as text.
A status pill on the attachment shows progress — Transcribing…, Transcribed, or Could not be transcribed. You can skip transcription per attachment, or retry a failed one. A message with pending audio cannot be sent until every attachment is either transcribed, skipped, or failed.
Admins must configure a provider model tagged transcription for this to work — see AI providers. Transcription calls are billed per minute of audio and recorded in the usage ledger alongside chat tokens.
Size and count limits
- Maximum file size: 100 MB per file by default. Admins can set a lower per-MIME-type cap (e.g. 25 MB for audio) in the Upload policy.
- Audio duration: audio and video uploads are capped at 4 hours of audio. Longer files are rejected on upload — split the recording into shorter segments.
- Maximum files per message: 10. For bulk ingestion use the knowledge base instead.
Where attachments live
Files attached to chat stay with the conversation — they’re not added to the shared knowledge base automatically. If you want the AI to remember a file for later conversations, upload it to the knowledge base separately.
Deleting a conversation also deletes its attachments unless your organisation’s retention policy keeps them longer.
Security
Uploads are scanned for viruses and blocked mime-types before they reach the model. If your admin has enabled PII detection, text extracted from attachments is run through the same rules as typed messages. Last modified on April 21, 2026
