Skip to main content

​
Why this matters

Some of your most useful information never makes it into searchable text: a screenshot of an error, a photo of a whiteboard, a chart embedded as an image, a scanned document. Chief reads images so that a picture you drop in doesn’t become a dead end. Attach one to a message and Chief looks at it along with your question. Add one to your Library and it becomes part of the same knowledge you query everything else against. The value is simple: fewer “I know it’s in there somewhere” moments, because image-based files can join the rest of your Project instead of sitting outside it.

​
How it actually works

Chief Vision Ingest - adding images via Upload files or Paste content
Where you add an image decides what Chief does with it:
  • Attached to a message. Paste an image into the prompt bar, drop it on the prompt bar, or pick it through Upload files in the prompt bar, and a PNG, JPEG, GIF, or WebP image up to 5 MB attaches to that message as a tile. Chief reads the image itself along with your question. The image stays with that chat only. It isn’t added to your Library unless you turn on the Library icon in the tile’s bottom-right corner before you send; that icon starts off for images.
  • Added to your Library. Upload an image with Upload files on the Library page, shown above, or drop it anywhere other than the prompt bar. Image files added this way are recognized as their own type and run through Chief’s ingestion pipeline rather than the document path. As part of that, Chief normalizes images so they’re cleanly readable by the underlying models — for example, re-encoding PNGs that arrive with non-standard metadata (a common quirk of screenshots taken on phones) so nothing trips up downstream processing. Once an image is ingested into a Project, it lives alongside your other files and can be brought into a task the same way. Image types that can’t be attached, such as TIFF or BMP, always take this path.
Images are part of Chief’s Universal Ingestion story — the same capability that takes in PDFs, audio, video, spreadsheets, and transcripts. The intent is that image-based documents stop being invisible and become queryable intelligence with no preprocessing on your side.
ℹ️ Good to Know: Chief reads what’s visible in an image — it can’t recover information that isn’t actually shown there. A blurry photo, a cropped chart, or a screenshot missing the relevant region limits what Chief can work with, the same way it would for a human reader.

​
A worked example

A teammate pastes a screenshot of a dashboard into the prompt bar instead of exporting the underlying numbers, and asks what changed this week. Rather than that image being a dead end, Chief reads it with the question and answers from what it shows — no re-typing. If the team will want that dashboard again in other chats, the teammate turns on the tile’s Library icon before sending, or uploads the screenshot to the Library instead, so it’s ingested into the Project alongside your documents and transcripts.

​
FAQ

The most common causes are image quality and framing: if the relevant text or chart is small, blurry, or cropped out of frame, Chief has less to work with. Re-uploading a clearer, fuller image usually helps. Two other causes come from where the image went. An image attached to a message is shown to Chief again for a limited number of later turns, so deep into a long chat Chief may no longer see it; attach it again. And an image attached in a different chat, with its Library icon off, never reached this chat or your Library, so attach it here or add it to your Library.
Yes. This article is about Chief taking in image files you provide. Generating new images from a prompt is a separate, beta capability.
It depends on where you add it. In the prompt bar, you can attach PNG, JPEG, GIF, and WebP images up to 5 MB, and up to 5 files per message. A larger image is refused with “Images can be up to 5 MB, and {name} is larger.” and isn’t added anywhere. Other image types dropped on the prompt bar go to your Library instead. The Library checks each file against a fixed list of types before uploading; PNG, JPEG, GIF, WebP, TIFF, and BMP are on it, but HEIC photos from an iPhone are not, so convert those to JPEG or PNG first. Learn more about supported file types.