Skip to Content
Offline GST invoice extraction

It reads your invoices. It never reaches the internet.

Accountant reads a folder of invoices: born-digital PDFs, flatbed scans, phone photos, thermal receipts. It produces validated, GST-compliant data ready for GSTR-1, GSTR-3B and Tally. One Windows installer. No cloud, no portal call, no telemetry.

Works with the network unplugged Unlimited invoices Installs in minutes
Batch summary September
INVOICES PROCESSED
  Pages read          1,000   24m 11s
  Posted without edit   871
  Sent to review        129
  Arithmetic corrected   17

CORRECTED  INV-4471  invoice total
  read       ₹1,10,000.00
  corrected  ₹1,18,000.00
  proof      1,00,000 + 9,000 + 9,000
  cause      8 read as 0, one repair fits

REVIEW  INV-4488  GSTIN check failed
REVIEW  INV-4502  CGST  SGST, intra-state

Nothing was sent anywhere.
< 0.1%
Silent-error rate
The number we commit to, not a headline percentage
25 min
Per 1,000 pages
On an ordinary office desktop, with no graphics card
13
Indian languages
Read on your own machine, with everything bundled
0
Data sent out
Watch the network while it runs. Nothing leaves.
The problem

An automated error is worse than a manual one, because nobody was watching when it was made.

39% of invoices carry a minor error: a typo, a transposed digit, a swapped tax head. Those errors surface months later as a GSTR-2B mismatch, when correcting them is expensive or impossible.

What every tool does

Automates the typing

Extraction is the easy half. Every product on the market will pull a number off a page and put it in a column for you.

What the rest skip

Telling you which row is wrong

Knowing which of the thousand rows it just produced is quietly wrong is the harder problem, and it is treated as an afterthought.

What we do

Fix it, or flag it honestly

An error you can see costs eight seconds. An error you cannot see costs a GST notice. Fewer than 0.1% of our fields are wrong and unflagged.

Why this one

Three things that make this different

Not positioning. Each one is something you can check for yourself in an afternoon.

It is genuinely offline

Not “offline-capable”. Not “works offline with reduced features”. The reading engine, the HSN master, the dated GST rate history and the e-invoice certificates all ship inside the installer.

Updates arrive as signed packs you carry in on any medium and import from Settings.

NO CLOUD NO TELEMETRY

It refuses to be silently wrong

A compliant e-invoice carries a signed QR code. We read it and verify the government signature on your machine, so those fields are taken exactly as issued. Digital PDFs are read directly, with no guessing at all.

Where the page has to be read from an image, two independent readers must agree. What cannot be proved gets flagged, never guessed.

SIGNED QR VERIFIED TWO READERS AGREE

It gets better at your vendors

Indian businesses receive repeat invoices from a small, stable vendor set. The first time you correct a field on an invoice from a given GSTIN, the app remembers how that supplier lays out their bills.

Every later invoice from them is read against that memory first. Week eight beats week one, and we measure it.

LEARNS PER SUPPLIER IMPROVES WITH USE
The part nobody else does

It corrects a misread digit instead of just complaining about it.

A GST invoice has to add up in several ways at once. Line items sum to the subtotal. Taxable value times rate equals tax. CGST equals SGST on an intra-state supply. The grand total reconciles within a half-rupee of round-off. When one of those stops holding, a single misread character is enough to explain it, and that is a solvable problem rather than a reason to shrug.

Worked example
Taxable value₹1,00,000.00
CGST @ 9%₹9,000.00
SGST @ 9%₹9,000.00
Invoice total, as read ₹1,10,000.00
Invoice total, corrected ₹1,18,000.00

1,00,000 + 9,000 + 9,000 = 1,18,000. An 8 was read as a 0. Exactly one change to one character makes the whole invoice add up, so it is applied and recorded with the working that proves it. Had two different corrections both worked, every field involved would have been sent to a human instead.

How it finds the fix

It tries every single-character mistake that reading a printed page is known to produce:

6 ↔ 8 1 ↔ 7 0 ↔ O 5 ↔ S

One correction that makes everything add up is applied and recorded. Two, or none, and the fields go to review. Nothing here depends on how confident the reading looked, which is the point: a confidence score cannot catch an error that looks confident.

Every amount is held as a whole number of paise and every rate as a whole number, from the moment it is read to the moment it reaches your return. Nothing drifts by a rupee on the way.

Verifiable, not asserted

Your clients’ data never leaves your office. Enforced five ways.

Offline is built into the product, not promised by a salesperson. Five separate safeguards stop a release going out if anyone, including us, ever changes that.

01

The build refuses to ship

Add anything that could contact an outside server and the release stops. Not a review step anyone can wave through.

02

The reader cannot connect

The part that reads your invoices has no way to reach anything beyond your machine. By design, not by setting.

03

Tested with the network off

Every release is tested end to end with networking off. If any part needed the internet, that test fails.

04

No updater, no check-ins

No automatic updater to hijack, and nothing checking in while you work. You choose when a version is installed.

05

Nothing can reach it

The product opens nothing for others to connect to. No firewall prompt, and no new way into the machine.

The pipeline

Pile of invoices in. Filing-ready workbook out.

Drag a folder in, or point it at a watched folder. Multi-page PDFs and ZIPs are unpacked for you. Progress is saved for every page, so closing the app mid-run picks up where it stopped.

Step 01

Sort

Every page is identified: tax invoice, credit note, debit note, delivery challan, or something that is not an invoice at all.

Step 02

Read

Digital PDFs are read directly. Scans and phone photos are straightened, cleaned and read twice, in 13 Indian languages.

Step 03

Extract

Every field a tax invoice must carry under Rule 46 is pulled, line by line, and any e-invoice QR code is read and verified.

Step 04

Check & correct

GSTIN validity, place of supply, the rate in force on the invoice date, the full rule catalogue, then the arithmetic correction.

Step 05

Review & export

Only what needs a human reaches the review screen, which is built for the keyboard. Median time on a flagged invoice: under 25 seconds.

What comes out

Everything the return actually asks for

Not a flat list of invoices for you to reshape. The data comes out already sorted into the tables the GST portal expects, with the summaries computed and the totals reconciled.

GSTR-1 Outward supplies
B2B invoices
B2C large, inter-state
B2C small, consolidated
Credit and debit notes
Exports and zero-rated
Nil-rated and exempt
HSN-wise summary
Documents issued
GSTR-3B Monthly summary
Outward supplies and tax
Reverse-charge inward
Eligible input tax credit
Ineligible credit, separated
Supplies to unregistered
Totals reconciled to GSTR-1
Tax payable by head
Cess kept separate
CHECKED Before it leaves
GSTIN validity, both parties
Place of supply
CGST and SGST checked apart
Rate in force on the invoice date
HSN against the rate charged
Reverse charge flagged
Invoice numbering gaps
Every figure traced to its image
Formats
GSTR-1 offline-utility workbook GSTR-3B summary Tally Prime XML CSV JSON

The workbook you upload to the portal, not a spreadsheet you have to rearrange first.

Tally, both ways

Export a file and import it, or switch on the live connector and post vouchers straight into a running Tally on the same machine.

The connector reads your ledgers and stock items back, so supplier and item mapping is suggested rather than typed. “Does it work with Tally?” is the first question we get, which is why this is in version one.

Provable, months later

Documents are stored encrypted on your own disk, and every action is written to a tamper-evident record that an auditor can be shown.

When a figure is questioned in November, you can show the original image, what was read, what was changed, by whom, and why.

Speed, on an ordinary machine

Measured on a four-core desktop with 8 GB and no graphics card.

InputPer page
Digital PDF0.15 s
Clean scan1.2 s
Phone photo3.0 s
1,000-page mixed batch≤ 25 min

What it needs

One installer. Nothing else to install first, and no IT department.

MinimumRecommended
Windows 10Windows 11
4 cores, 8 GB RAM8 cores, 16 GB RAM
6 GB free disk20 GB free disk
No graphics card neededUsed if present
Licensing

Per seat. Unlimited invoices. No meter.

Per-invoice pricing punishes you for processing everything, which is exactly the behaviour that makes the tool worth having. So there is no per-invoice fee. One annual number, compared against one junior’s salary.

Standard
Request pricing
annual, 1 seat

An SME accounts clerk, running one company’s books.

The full pipeline, the full rule catalogue, Tally and the return workbooks.

Most practices
Professional
Request pricing
annual, 1 seat

A small CA practice, carrying the risk on every return it signs.

Adds multi-company, template export and the auditor pack.

Firm
Request pricing
annual, 5 or 10 seats

A CA firm, with several people working the same client list.

Shared template packs, so what one seat learns benefits everyone.

Prefer to own it? A perpetual licence is available in place of the annual one, with yearly maintenance. Extra seats, data migration and on-site training are quoted alongside.

Request pricing
Objections

The questions we actually get

“Offline means no updates.”
Signed packs, imported in one click. The version and its age are shown permanently, so you always know how current your rate tables are. And no vendor outage can stop you filing.
“Cloud tools are cheaper.”
Work out their per-invoice rate at your real volume. At month-end they are not. And your client data is sitting on somebody else’s server either way.
“What if you go out of business?”
The software keeps working. It has no dependency on us, no server to switch off, no licence to re-check. That is not true of any cloud competitor.
“Can it do 100%?”
No, and nobody can. What it does is tell you which ones to check, and be right about that. We will show you our silent-error rate rather than a headline accuracy number.
“My juniors do this fine.”
They do. The question is what their week costs, and what a reconciliation mismatch costs when it surfaces in November.
“Does it work with Tally?”
Both ways. A file you import, or vouchers posted directly into a running Tally on the same machine, with your ledgers read back for mapping.

Give us fifty of your own invoices and ninety seconds.

Drop them in. Watch them process. Then open the one it corrected and see the arithmetic that proves it. Then we unplug the network cable and do it again.