Multimodal input

Screenshots, photos, and video clips go straight into your prompt — no upload step, no special syntax.

buildwithnexus multimodal session: a clipboard screenshot pasted with Ctrl+V renders as an inline thumbnail in the transcript and is diagnosed, a video is parsed with ffmpeg into 8 sampled frames with duration, resolution and fps metadata, and the findings are written to a .docx report whose path opens in Word on click — A pasted screenshot (`Ctrl`+`V`) rendered inline in the transcript, an `@demo.mp4` attachment sampled by ffmpeg with timeline metadata, and the findings written to a `.docx` — all in one session.

Paste an image

Ctrl+V grabs the image on your system clipboard (Wayland, X11, macOS, WSL), saves it to a temp file, and attaches it as an @path token. If the clipboard holds text, it inserts at the cursor instead.

See what you attached

Attached images render inside the transcript as truecolor half-block thumbnails (up to 64×36 px, decoded via ffmpeg) — you can confirm the right screenshot went to the model without leaving the terminal. Video attachments preview the same way. Requires a truecolor terminal; on anything less the preview is skipped, never garbled.

Attach files

> why does the header overlap? @screenshot.png
> summarize what happens in @demo.mp4

Bare image paths work without the @ too. Supported images: png, jpg, gif, webp.

Video via ffmpeg

Video attachments (mp4, mov, webm, mkv, avi, m4v, …) are probed with ffprobe and sampled with ffmpeg: up to 8 evenly-spaced frames (scaled to ≤768px) attach in order, with a metadata block — duration, resolution, fps — so the model understands the clip's timeline. Requires ffmpeg on your PATH.

Word documents out

Ask for a shareable report and the agent writes a real .docx with the built-in create_docx tool. The path in the ✦ create docx line is an OSC 8 file:// hyperlink — on supporting terminals (iTerm2, WezTerm, kitty, Windows Terminal, GNOME Terminal, …) clicking it opens the document straight in Word or your OS default app.

Vision gating

Attachments only ship to models that can see them. Text-only models print ⚠ current model is not multimodal — image not attached instead of silently dropping your screenshot. Detection details in Models & providers.