Getting things in
Drop in PDFs, Office documents, images or a folder. They are converted to text before they cost you anything, so a 40-page PDF is not 40 pages of tokens in four seats at once. Paste an image straight into the composer where the model supports vision.
Where the deployment has web search configured, a question can carry search results that every seat reads.
Getting the result out
- Copy any message as the model wrote it, markdown and all.
- Export the transcript as Markdown, or the agreed answer as HTML, PDF (through your browser's print dialog) or a Word document.
- Reports — an answer with tables becomes a page where every row has a done box and a note, shared by link. For the answers that are really a checklist.
- Share a discussion read-only with a link. Revoking it is permanent.
- Read aloud, with a recognisable voice per seat, and a voice mode where the microphone listens, a pause sends, and the reply comes back as speech.
Changing your mind
Edit a question and ask again, regenerate an answer, or branch the discussion into a new one up to a chosen message. The first two genuinely remove what came after — the cut is recorded, so another device cannot push the withdrawn turns back.
Finding things again
Search across every discussion, or ask a question about them — "which discussion was the one about the Paddle webhook?" Pin an answer to keep it at the top of its discussion on every device, and star a discussion to float it above the date groups.