A brief says what you want. An attachment shows it. Drop a screenshot of a layout you like, a PDF of brand guidelines, or a screen recording of an interaction, and the agent works from the real thing instead of your description of it.
Three ways to attach
- The + button in the composer opens a file picker.
- ⌘V (Ctrl+V on Windows) pastes files or an image from the clipboard.
- Drop files anywhere on the chat panel.
Each file shows up as a chip above the composer. Clicking a chip before you send removes it. When you send, the files go with that turn.
Where attachments go
Attachments are copied into your project folder under
.1design/attachments/, with a short timestamp added to the name so two files
called screenshot.png never collide. The turn's prompt names those paths, and
image files are passed to the agent as images when its runtime supports them.
Nothing is uploaded anywhere by the studio itself. The agent CLI you use reads the files from disk the same way it reads the rest of the project.
A file that can't be written never turns into a chip. You get an error card that says why instead of a turn that silently went out without it.
Reopen what you already sent
Every sent message lists the files it carried as filename chips. Click one to open it:
- Images open in a preview dialog inside the studio, with the filename, Close, and Open in default app.
- Other files (PDFs, code, exported tokens) show the same dialog with Open in default app, which hands the file to whatever your system uses for that type.
This works on conversations saved before 1.9.0 too, so an old turn's references are one click away when you pick a project back up.
Video references
A screen recording is often the fastest way to describe motion or an interaction. Attach a video file the same way as any other attachment, up to 250 MB per clip. A larger clip is refused by name with a hint to trim or compress it, rather than dropped.
Videos get their own chip, and the chip shows where the clip is in processing: Reading video, Extracting frames, Analyzing reference, then Ready. If something needs your call it shows Needs input, and if processing fails it shows Could not process.
Click a video chip to open its detail card: the extracted storyboard frames and what the reference covers, plus the actions that apply to its state:
- Range: process only part of the clip.
- Analyze with…: run a separate pass on a vision-capable runtime that writes a cited brief your current model can read. Use it when your agent can't look at images itself.
- Continue text-only: send the turn without visuals. The run report
records
visual-reference: unverifiedinstead of guessing at what the clip showed. - Retry or Cancel while it's processing, and Remove to take the chip off. Removing a chip doesn't delete frames an earlier turn already used.
Attachments on queued and steered messages
A follow-up you queue while a turn is running keeps its own attachments. Editing the queued text doesn't change them, and applying it early with Steer now sends that message's files, not whatever is in the composer at the time. See Write a brief that works for Continue, Steer and the queue.
Tip: An attachment works best with one sentence that says what to take from it: "match the spacing of this card, not the colors" gets a better result than the image on its own.