How it works
Neo automatically detects image attachments on a ticket and runs each one through a vision model. The model produces a description of what the image contains, and that description is added to the ticket context the agent works from. From there it’s treated like any other ticket content — the agent can pull a figure off a spreadsheet screenshot, read an error message from a screen capture, or factor a photo into its diagnosis and dispatch decision. This is built in, not a tool you enable. There’s no “read image” or “OCR” entry in the agent’s tool list and nothing to configure — image reading happens as part of loading the ticket, so any agent that reads the full ticket gets the image content automatically, and so does the Ticket Triage action. A description is kept for 30 days once an image has been read, so the same image file (a signature logo, say) is described once and reused across the tickets and runs it appears on. The match is on the file’s bytes: a copy that a mail gateway re-encoded is read again. (Because it isn’t a tool, you won’t find it on the Tools Overview page — it’s a capability of how Neo reads tickets.)What’s supported
Good to know
- Neo reads the picture’s content, not a guaranteed pixel-perfect transcription. For a spreadsheet screenshot Neo will read the visible figures and labels; for a very dense or low-resolution capture, exporting the underlying data (e.g. a CSV) is still the most reliable way to get every value across.
- Non-image documents. Email files (
.eml) and PDFs attached to a ticket are read the same automatic way — see Reading Email and PDF Attachments. Other document types (Office files) are not read from ticket attachments; this capability is specifically for raster images (PNG / JPEG). - An image Neo cannot read is marked, not hidden. When Neo cannot download an image that a note references, the note shows
[Image could not be read here - its content is unknown unless an image description below this text shows it]in its place. The agent can then tell that it did not see that image there. The other images in the same note are still read. - Spacers and blank images are skipped. Email signatures carry thin dividers, spacers and tracking pixels with nothing to read. Neo does not send an image under 8 px on a side, or a small image of one colour, to the vision model: the ticket shows a short note such as
[Decorative image of 1 x 26 px (a spacer, divider or tracking pixel) - no content to describe]in its place. - It’s about images on tickets. This is the supported path for “someone sent in a screenshot”: attach it to the ticket and Neo reads it. Image content pasted into other surfaces isn’t read the same way — keep the image on the ticket so the agent can act on it.