Nexora
Menu
ខ្មែរEnglishkm
Start free trialSign in
NexoraOCR by Nexora

NexoraOCR is Nexora's document product.

Send a document to a chat. Get it back as a file you can edit.

NexoraOCR reads photos and PDFs — Khmer first — and returns an editable Word or Excel file, or the fields checked and ready to use. It works inside Telegram, so there is nothing for your staff to install.

Book a demo with your documentsRegister your interest
The Mini App, inside Telegram. Two modes, one document at a time.
Where it runs

Inside Telegram, where your team already is.

NexoraOCR is a Telegram Mini App. Your staff open a chat they already have, send the document, and the answer comes back to the same conversation.

  • Nothing to install, no software to roll out, no laptop needed.
  • No account for staff to create — Telegram is the identity.
  • Works from a phone in a warehouse or at a counter, on a photo taken there and then.
  • The finished Word or Excel file is delivered back into the chat, ready to forward.
  • A browser console exists too, for developers who want the API, keys and webhooks.
Upload, convert, send. The file arrives in the same conversation.
Two modes

Convert the file, or extract the data.

You choose before you send, because they answer different questions. One gives you a document to keep writing in; the other gives you fields to check and use.

Convert file

A photo or PDF comes back as Word or Excel, laid out like the original.

  • Word (.docx): an editable document with headings, paragraphs and tables you can keep writing in.
  • Excel (.xlsx): the tables and figures from the document, as a spreadsheet.
  • The layout is rebuilt, not screenshotted — you can edit every line.
  • A photographed page is wrapped into a page first, so a phone photo takes the same route as a scan.

Extract data

The document becomes named fields and line items, with a check on the numbers.

  • Vendor, invoice number, dates, tax ID, currency, totals and line items.
  • Text, JSON, CSV, Markdown, Word or Excel from the same result.
  • Templates for the document types you handle repeatedly.
  • An assistant that can summarise, translate or answer a question about the page.
Checked, not guessed

It re-adds your invoice and tells you when it does not add up.

Extraction that hands back a total nobody verified is how a wrong number gets into the books faster than typing would have put it there.

  • Line items are summed and compared with the stated total.
  • Subtotal plus tax is reconciled against the total.
  • A due date before the invoice date, an implausible date, an unusually large quantity — each is flagged in plain words.
  • An amount that could be read two ways is marked ambiguous rather than picked for you.
  • Every field sits beside the original page, so a person confirms in seconds.
Extract data: the fields, and the arithmetic checked against them.
Khmer, properly

Built for the script most OCR tools drop.

Khmer is a low-resource script with stacked subscripts and no word spaces. General-purpose OCR loses it quietly — a page comes back shorter than it went in, and nothing says so. That failure is the reason this product exists.

  • Khmer is the case we tune and measure on, not a language we happen to accept.
  • On our own PDF corpus the Word rebuild keeps 97.8% of the Khmer text — measured, not estimated.
  • Mixed Khmer and English on one page is the normal case here, and is handled as such.
  • Handwritten Khmer is the known limit: our production model is not reliable on it, so the system refuses rather than guessing.
  • Interface in ខ្មែរ and English.
What it will not do

The three answers that make the rest trustworthy.

A document tool is only useful if you can believe it when it says it worked.

It says when it cannot read
If the text cannot be read reliably, no file is made and the app says so. We would rather tell you than send you a guess.
You are not charged for our failure
A document that fails is not counted against your plan. The charge is the last step of a successful run, not the first step of an attempt.
It never claims delivery it cannot see
When the bot has been asked to send your file, the app says it is on its way — never that it arrived, because only your chat knows that.
A refusal is a correct outcome with its own screen, not an error. Nothing is invented, and nothing is charged.
Who it is for

Teams that still key documents by hand.

Accounting and admin teams

SMEs, accounting firms, distributors, NGOs and clinics entering supplier invoices and receipts.

Operations and field teams

Delivery notes, forms and paperwork photographed where it is signed, not carried back to a desk.

Software teams

An extraction API, keys and webhooks, without building the pipeline.

Documents it handles
Supplier invoicesSales receiptsDelivery notesPurchase ordersExpense receiptsReports and lettersFormsScanned pages in KhmerPhotographed documents

Printed and photographed pages, in Khmer, English or both. Handwriting is not reliable yet and we say so rather than letting a demo fail. Bring three of your own documents and we will run them in front of you.

For developers

An API, not just an app.

The same pipeline behind the Mini App is available to your own systems, through a console built for that audience.

  • API keys issued and revoked from a browser console.
  • Webhooks when a document finishes, with delivery attempts recorded.
  • Structured results in a consistent shape, whatever the document.
  • Templates so a recurring document type extracts the same fields every time.
Your documents

Where they go, and for how long.

Processed, then retained by your setting
Documents are processed on Nexora's servers and kept only as long as your retention setting allows.
Demo uploads are deleted
Anything you send us for a demo is used for that demo and deleted afterwards, within 14 days at the latest.
Separated per customer
Your documents and results are not visible to another customer.

Questions teams ask

Does it read Khmer?
Yes, and it is the case we build for. Khmer stacks subscripts and does not space its words, which is exactly where general OCR tools quietly lose text. On our own PDF corpus the Word rebuild keeps 97.8% of the Khmer on the page. Mixed Khmer and English is the normal case here and is handled as one document.
Do we have to install anything?
No. It runs as a Telegram Mini App, so your staff open a chat they already use. Developers who want the API get a browser console instead.
What comes back — text, or a real file?
Either. Convert mode returns an editable Word or Excel file rebuilt from the page. Extract mode returns named fields and line items, which you can also take as text, JSON, CSV or Markdown.
How accurate is it?
We do not quote a single number, because accuracy depends on your documents. We measure it on yours during the demo. What we can promise is that the numbers are checked against each other, and that the system refuses a page it cannot read rather than inventing one.
What about handwriting?
Handwritten Khmer is not reliable on our production model — we tested a model that reads it and rejected it as far too slow and expensive for everyday use. The system detects the failure and abstains instead of returning a guess.
Where do my documents go?
To Nexora's servers for processing, then kept only as long as your retention setting says. Documents you send for a demo are deleted afterwards, within 14 days at the latest.
Can it feed my accounting system?
Export to Excel or CSV works with any system today, and the API plus webhooks let a developer wire it directly. Direct connectors are built per customer rather than claimed as ready.
How do I try it?
Book a demo and bring three of your real documents — ideally your hardest ones. We run them live and you see exactly what comes back.

See it running on your own clinic's day.

Start a free trial, or book a 30-minute demo in Khmer or English. No card, no commitment.

Start free trialBook a demo
Start free trial