Products / IRIS Vision
S.A.L.T CI IRIS Vision
The eye of S.A.L.T CI. Point a camera at a name card, a whiteboard, a document or a product, and IRIS reads it, understands it and reports on it, on local vision models, on the organisation's own servers.
Senior Engineer
Contoh Engineering Sdn Bhd
weihan@contoh.my
So much work starts with something on paper
Name cards from a meeting, a whiteboard full of decisions, a scanned form, a photo of a nameplate on a machine. Someone usually types it all in again by hand.
Retyped by hand
Cards, forms and boards are copied into systems by people, slowly and with errors.
Sent to foreign clouds
Popular vision services upload every photo, including identity documents and internal papers, to someone else's servers.
One tool per job
A card scanner, an OCR tool, a chart reader: separate apps that do not feed the systems people use.
IRIS in one line
One vision service that reads documents, scenes, cards, boards and objects on the organisation's own servers, used directly as an app and called by every other S.A.L.T CI application as its eye.
Two ways to use IRIS
Point, capture, understand
- A camera-ready web app installable on any phone
- Choose a category or let IRIS decide
- Get a structured result and a written report
- Save, share or send it on
The eye of every application
- MOM Studio: name cards to attendance, whiteboards to minutes
- TERAS: pictures in chat and documents
- PANGLIMA: understanding what is on screen
- Any app, through one API
Thirteen categories, one eye
IRIS classifies each image, applies the right reading for its kind, and returns structured fields as well as a plain-language description.
Business cards
Names, titles, organisations and contact details as fields.
Documents and forms
Text, sections, tables and form fields from printed pages.
Whiteboards and diagrams
Notes, decisions, lists and arrows from boards and flip charts.
Charts and data
Values and trends read from charts and printed tables.
Products and packaging
Product identity, labels and multilingual packaging text.
Equipment
Nameplates, models, serial numbers and visible condition.
Damage and defects
Visible damage described for inspection and claims.
IDs and certificates
Fields read for verification workflows, with personal data handled on site.
Scenes and places
What is in a scene and where, for site and field reports.
Vehicles
Type, markings and visible details.
Screens
What is on a captured screen, for desktop assistance.
Signs and notices
Text in Malay, English, Chinese and Arabic scripts.
Plus a general category for anything else. Accuracy depends on the vision model chosen; larger models read dense handwriting better.
Photos arrive the right way up
Phone photos carry their rotation in hidden metadata. IRIS straightens every image before reading it, so a sideways whiteboard is read as well as an upright one.
Capture, classify, read, report
Every image passes through the same five-stage pipeline. Each stage can be tuned, and each runs on the organisation's own models.
Camera, upload or another app; straightened and resized
Which of the categories this image belongs to
A local vision model with a prompt for that category
Optional: context from the knowledge base or permitted web
Structured fields and a written report
Two reading modes
Identify and research
A detailed reading with identification and research synthesis, for products, equipment and scenes where context matters.
Lean and fast
A short prompt for whiteboards, name cards and pages, used by MOM Studio so meetings are processed quickly.
Output
Four languages
Results in Bahasa Melayu, English, Mandarin or Arabic.
Batches
Several images in one request, such as all the cards from a meeting.
Reports
A written report per image or per batch, ready to save or attach.
Configurable limits
Image size, quality, timeouts and the length of each answer are settings, so an administrator can trade speed for detail per deployment.
The eye every application shares
IRIS is built once as an MCP server. Improve the eye, and every application that reads an image gets better the same day.
| Application | What IRIS does for it |
|---|---|
| MOM Studio | Name cards into the attendance list; whiteboards, slides and printed papers into the minutes, in meeting mode |
| TERAS | Pictures attached to chat, and scanned or handwritten pages in knowledge collections |
| PANGLIMA | Screen awareness for the desktop companion: what the person is working on |
| TERUSAN | Reading supporting documents attached to a case, at the AI autonomy level the step allows |
| SENTINEL and TETAMU | Scenes and documents where a still image needs a written description |
| Any application | One API for capture, classification, reading and reports |
Example: a meeting, end to end
Nine name cards and two whiteboards
Cards to fields; boards to notes
Merged with the transcript
Attendance table and decisions filled in
A photo never blocks the work
If one image cannot be read, the others still go through. The calling application gets a clear warning naming the image instead of a failure.
Photos of people and papers stay on site
Identity cards, internal papers and faces are sensitive. IRIS runs local vision models through the S.A.L.T CI core, and external providers are refused unless the organisation explicitly allows them.
Local by default
Vision runs on models served by TERAS on the organisation's own hardware, with no API keys and no external calls.
External providers refused
A cloud vision model can only be used when an administrator enables it; otherwise IRIS refuses to start with one.
Platform identity
Sign-in, roles and audit through the S.A.L.T CI core.
Choose the model
Small models for cards and pages; larger ones for handwriting and dense scenes. Switching is a setting.
Any hardware
A container beside TERAS on servers, VMs or private cloud; GPU recommended for vision models.
Compare before relying
Try models and prompts side by side on real photos in TERAS Model Studio.
Compared with common options
| Capability | IRIS | Cloud vision APIs | Phone scanner apps |
|---|---|---|---|
| Images stay on your servers | Yes | No | Varies |
| Runs fully air-gapped | Yes | No | Partly |
| Cards, boards, documents and objects in one service | Yes | Separate APIs | Usually one kind |
| Feeds minutes, chat and workflows directly | Yes, as an MCP server | Needs integration | Export only |
| Malay and local scripts | Yes | Varies | Varies |
Categories describe typical offerings; individual products differ.
Part of one sovereign platform
IRIS Vision runs on the TERAS core with the same sign-in, audit and models as every other S.A.L.T CI application.