Visual case file and unified knowledge search
Turn the photos a customer already sends into a case file with roles, captions, marks and sealed evidence; read a photo with AI as a suggestion, never a verdict; record whether the case was solved remotely; and search your knowledge base, your external sources, your internal documents and your resolved cases in one place, with the origin shown on every result.
What this is, and what it is not
This is not live video. There is no call button, no room and no screen sharing: interactive video needs a real time media provider that SellioCRM does not have, and a button that opened an empty screen would be worse than no button at all. What this does is the half that works with what you already have: the photo the customer sent over WhatsApp, the file that arrived by email, the picture the technician took in the app.
The visual case file
On the case, every image gets a **role**: problem, before, after, solution or reference. A file nobody classified stays as reference, because pretending it proves something would be deciding for the agent. Images with a before and an after are paired, and a lone "before" is never shown as a pair.
You can **mark the image**: drag to draw a box, click to drop a point, and write what it is. The mark is stored beside the file in relative coordinates, so it lands on the right spot on a big screen, on a phone and in a report. **The file itself is never touched**: burning an arrow into the JPEG would destroy the original, and a destroyed original stops being evidence.
Sealed evidence
Keeping a file is not proving anything. **Seal** reads the bytes, records a sha256 fingerprint together with the size, the name, the path and the channel the image came from, and freezes that in a line nobody can edit. **Verify** reads the bytes again and compares: `ok` when they match, `tampered` when the file changed, `missing` when it is no longer in storage. Those are three different facts and the product says which one happened. Sealing twice does not create a second moment of proof, and sealing a file the system cannot read is refused.
Reading a photo with AI
When system AI is available, **Analyse** sends the image with the case and the equipment as context and returns what it sees, the likely causes with a confidence, parts that may be needed, and an answer to the question dispatch actually asks: does this need an on-site visit? The answer can be **yes, no or "cannot tell from the photo"**, and the third one is a real answer. The result is a suggestion, not a technical report, and the screen says so. It is a human action, not a sweep: nothing is analysed unless someone asks, and the same image is never charged twice.
Solved remotely, and first time fix
On the case you can record whether it ended **remotely** or **on site**. That one click is what makes "how many visits did remote support avoid" answerable. "Not sure yet" is a real option: the report separates what was declared from what nobody declared and shows the coverage next to the rate, so a high rate with two declared cases does not look like a victory.
First time fix comes from field work: a job counts as a first time fix when no return job hits the same equipment (or the same site, when there is no equipment) within a window, and the window travels with the number so two screens never compare different rulers.
Knowledge sources and one search
Under Settings, **Knowledge sources** is where you point at what the base should also read: an address (one page), a sitemap (a whole manual, in batches) or your internal document repository. The pages are fetched on a schedule and re-fetching costs nothing when the content did not change.
An external document **does not become an article**: your base stays what your team wrote and reviewed. Audience is set per source: a public manual can appear on the portal, an internal runbook never does.
The unified search then reads four origins at once: your knowledge base, your external sources, your internal documents and your resolved cases. Every hit carries **where it came from**, and the footer says which origins were not consulted and why, because an empty list would otherwise read as "nothing exists about this" when the truth is "we did not ask there".
What is deliberately missing
SharePoint, Confluence, Notion and Drive are not offered. Each one needs its own OAuth application and consent; offering the field and failing later would be worse than not offering it.