ARGUS
Investigation intelligence
Product tour
Argus AI

Every feature,
working on screen

Not a list of claims. Each capability below runs as an animation of the thing itself — the report being scanned, the fields flying into the case, the sections lighting up, the search being understood. Press replay on any of them.

Runs on your server No case data leaves the building Same answer every time
One platform, every jurisdiction
01 · Automatic FIR reading

The report reads itself

Drop the first information report into the create-case box. Text is lifted in the browser — a scanned report never crosses the network — and between twenty and thirty labelled particulars are pulled out along with every identifier in the narrative body.

Numbers are attributed to a role as they are extracted, so an action aimed at the accused is never aimed at the victim.

FIRST INFORMATION REPORT
FIR / Crime88 / 2026read
Police stationMadurai Cyber Crimeread
SectionsBNS 318(4), 319(2), 61(2)parsed
Complainant80 yrs · 1 contact numberrole: victim
Accused4 WhatsApp numbersrole: accused
Accounts5 · 3 opened by studentsflagged
Total lost₹53,00,000reconciled
Replay
Indian script detected
புகப்பாளர் முக்கிய புகப்பின் சாரம்
நிகழ்வு நாள் : 12.02.2026
காவல் நிலையம் : மதுரை
read
Script detectedTamil · Noto subsetok
Text layercorrupt — sent to OCRrecovered
Occurrence date12 Feb 2026parsed
Police stationMaduraiparsed
Replay
02 · Language

Any Indian script, read and rendered

The script is detected before anything is parsed. A Tamil report whose embedded text layer is corrupt — a common failure when the font was subsetted badly — is caught automatically and sent to optical character recognition rather than being read as garbage.

The particulars then come back in English while the original stays exactly as filed, so nothing in the record is altered.

03 · Case identification

It reads the offence, then names the sections

The narrative is matched against the offence definitions, and the sections that follow from it are proposed — under the Bharatiya Nyaya Sanhita and, where the conduct is electronic, the Information Technology Act.

Proposed, not applied. The investigating officer accepts, removes or adds; what Argus removes is the blank page.

The callers represented themselves as officers of a central agency, confined the complainant to her home for five days by continuous video call, and by deceiving her into believing a warrant existed induced her to transfer ₹53,00,000 across five transactions to accounts opened in the names of students for a monthly consideration.
BNS 318(4)Cheating
BNS 319(2)Cheating by personation
BNS 61(2)Criminal conspiracy
BNS 336(3)Forgery
IT Act 66DImpersonation by computer
Replay
number 9840 71•••••
time night hours
place Coimbatore
period March 2026
duration > 5 min
03 Mar9840 71••• → 9500 22•••07:12
11 Mar9840 71••• → 9500 22•••12:48
11 Mar9840 71••• → 8870 41•••06:03
19 Mar9840 71••• → 9500 22•••22:31
Replay
04 · Natural-language search

Ask it the way you would say it

One box across the whole case. The question is separated into a number, a time filter, a place, a period and a minimum duration — and Argus prints how it understood you before it answers.

The parser is rules, not a language model, so it runs on an air-gapped server and gives the same answer every time it is asked. For output that goes into a case diary, repeatability beats eloquence.

05 · AI Findings

Fourteen to seventeen observations, each with its evidence

Not a summary. Specific claims — the rented mule layer, the absence of any GSM contact, the five-hour session covering the transfer window — each one clickable through to the row it came from.

Which is what makes a finding usable in a case diary rather than merely interesting.

Rented mule layerhighThree beneficiary accounts opened within eleven days, all by students of one college.
No GSM contact at allhighThe accused numbers made no calls to the complainant. Contact was entirely over WhatsApp.
Complaint names one, pattern implicates severalmediumThe complaint names a single student; the transactions span three accounts.
Five-hour session on transfer dayhighA continuous WhatsApp session covering the window in which the first transfer left the account.
Terminal beneficiary identifiedact nowThe trail stops at one account. Marked for freeze.
Replay
Preserve before the window closesNotice to the operator against the accused-facing numbers only — never the complainant’s.
Requisition CDR and IPDR separatelyWith the exact date range. A generic notice returns only the first.
Serve the bank on the beneficiaryAnd obtain the account-opening file, which is where a rented account shows itself.
Resolve the operator after portabilityBefore the notice is drafted, or it reaches the wrong nodal officer.
Freeze the terminal beneficiaryLayer-one money is recoverable; money at the cash-out is not.
Replay
06 · AI Direction

And then it says what to do next

The findings say what is there. The direction says what to do about it — eight prioritised steps in the order they should be taken, each naming the record it needs and what it is expected to prove.

A junior officer can work the list without a briefing, and a supervisor can see at a glance whether the obvious step was taken.

A multi-source case, worked the usual way
14 working days
The same case in Argus
3 working days
Roughly 0 days returned to the investigation — almost all of it from the first week, which normally goes on reshaping four operators’ exports before anything can be read.
Replay
07 · What it gives back

The first week is the one you get back

Nothing above changes what a record contains. What changes is how long it takes to become readable — and on a multi-source case almost the whole saving comes from the days that normally go on making four operators’ exports line up before any analysis can start.

Against a two-year retention window and money that leaves a layer-one account in hours, days are the currency that matters.

Comparison against a conventional single-source workflow on a case carrying a first information report, two operators’ call records, an internet detail record, one phone extraction and a bank statement. Your own figures will vary with the case.

See it end to end

One case, from paper to court report

Every capability on this page, in the order an actual investigation meets them — as one continuous sequence.

The case flow
Paper → extraction → findings → records → direction → report
Walk through it