The platforms that genuinely solve this assume an eight-figure budget and an implementation team. No Indian private-investigation firm, KYC desk or district cyber cell can buy one — so they fall back to free lookups, browser tabs and spreadsheets.
Investigation intelligence · Built in India
One detail.
The whole picture.
We build Agent 47. You give it one detail about a person — a phone number, a wallet address, a face, a document — and it works out the rest, then hands back a dossier with a source behind every line.
- 8 Disciplines
- 86+ Collectors
- 1,770 Curated sources
- 30+ Forensic engines
Home / 02 — The problem
The good software was
never priced for India.
So a cybercrime case still gets worked with a browser, a notebook and a spreadsheet.
24.3 lakhcybercrime complaints filed in India in 2025NCRP / I4C
₹22,495 Crreported losses to cyber fraud, 2025Ministry of Home Affairs
459dedicated cyber police stations for all of itBPR&D / MHA, 2025
Mapping one organisation's digital footprint by hand takes 40–60 analyst hours. In compliance the arithmetic is worse: the overwhelming majority of AML and sanctions alerts are false positives, and every one of them is cleared by a person, by hand.
The current wave of AI investigation tools will report a sanctions check as clear when the check never ran, and cite sources that do not exist. In a background check or a bail hearing, a confident wrong answer is a liability, not a shortcut.
Every case starts from zero. The same mule account, the same fake identity, the same shell director appears across fifty separate matters in fifty separate firms — and nobody sees the pattern, because no product is built to remember across cases.
Sources: Ministry of Home Affairs / I4C — NCRP annual figures 2025 · BPR&D infrastructure data, 2025 · published industry AML false-positive benchmarks, 2025–26.
Home / 03 — What we build
One product,
built properly.
Agent 47 belongs on every investigator's desk in India.
Agent 47 is an autonomous investigation platform. It takes a single seed — a name, a phone number, an email, a username, a face, a crypto wallet, a company or a document — and works out the rest on its own: planning the investigation, collecting across sources, validating what comes back, resolving identities, reasoning over the result, and stopping when the next action is no longer worth its cost.
What comes out is one evidence store, projected into a client report, an internal memo, or a court evidence pack with statutory references, PII masking and compliant access logs. Every sentence in it is bound to a stored artefact. Uncited claims are stripped before the report is written — they cannot survive to the page.
It investigates, rather than searching
A tool-calling agent that forms competing hypotheses, tests them, notices its own blind spots on a scheduled skeptic pass, and stops on its own judgement. Not a fixed script, and not a chat window.
The analyst stays in charge
The investigation is drawn as a plan — objectives, phases, decision branches, evidence rules. The agent reads that plan on every turn and must either follow it or formally log a deviation.
Web intelligence and file forensics, one box
Two separate industries with separate tools. Here they run into the same evidence graph — so a document's contents and its provenance are part of the same case, not two reports someone has to reconcile.
Home / 04 — The honesty contract
It cannot lie
about what it checked.
Every source returns one of exactly three states. A real result. A checked-and-genuinely-empty result. Or unavailable — blocked by a sign-in wall, a bot challenge, a rate limit, a paywall.
| Source | State |
|---|---|
| sanctions · watchlists412 ms | Result |
| corporate registry1180 ms | Result |
| breach corpus264 ms | Checked — empty |
| social platform0 ms | Blocked — sign-in wall |
| chain analysis2740 ms | Result |
| court records903 ms | Checked — empty |
| any source | Clear — never ranState does not exist |
A blocked source is explicitly not evidence of absence. The distinction is enforced in the architecture, not promised in a policy document — which is the only version of it that survives contact with a deadline.
Home / 05 — Why it holds
Architecture,
not features.
Which is why they would take years to copy.
02 — It remembers across cases
Fifty matters, fifty firms, one account.
A mule account or a shell director appears across dozens of unrelated matters. Agent 47 pools evidence across every case a firm holds, and surfaces the shared node that no single case could ever see.
It compounds in one direction only: every case a customer runs makes the next one shorter. That is also why the product gets harder to leave the longer it is used — the graph is theirs, and it is worth more every month.
03 — Forensics lives in the same box
One evidence spine, from the first artefact to the final page.
Every raw artefact is hashed into a custody chain the moment it lands. Certified forensic engines — hashing, metadata, error-level analysis, steganography, carving, OCR — run inside hardened sandboxed cells and write into the same graph as the web evidence.
An analyst can retract a wrong entity without deleting the evidence behind it, because the record of what was collected and the conclusion drawn from it are two different things.
If information cannot be verified, it is not presented as a fact.
Home / 06 — Where this goes
The same engine, pushed
deeper into the same problem.
On top of code that already runs.
Investigations
Police · investigation firms · KYC desks
One seed to a sourced, court-usable dossier. This is the product in the field today, and where the first revenue comes from.
Tuned editions
Banks · law firms · state agencies
Perpetual monitoring, mule-ring detection and ownership unwind for compliance. Document review and contradiction detection for legal. An air-gapped build for central agencies. Same engine, different defaults.
Pointed inward
Public figures · founders · executives
The same collection engine aimed at a consenting subject rather than an adversary: everything published about a person, impersonation accounts and deepfakes flagged as they spread, takedowns driven from one place.
Prevention
Governments · banks · citizens
Everything learned from investigating fraud, turned back on preventing it — criminals tracked across jurisdictions, identity fraud caught at onboarding, citizens warned before a scam lands rather than after.
None of this is a second codebase.
Signed · ActiveA signed Expression of Interest with the Cyber Police Station, Nashik Rural, Maharashtra, dated April 2026. Under active evaluation.Read the case note
Home / 08 — Questions
What people ask
before the first demo.
What does CrypsisX do?
CrypsisX builds Agent 47, an autonomous investigation platform. You give it one seed — a phone number, an email, a username, a face, a crypto wallet, a company or a document — and it plans the investigation, collects across 1,770 curated sources, validates what comes back, resolves identities, and hands back a dossier with a source behind every line.
How is it different from link-analysis tools or a generic LLM agent?
Link-analysis tools wait for an analyst to run each transform; generic LLM agents will report a check as clear when the check never ran. Agent 47 runs the investigation loop itself, and its architecture makes an un-run check impossible to present as a clear result. Web intelligence and file forensics also share one evidence graph instead of being two products.
Can the output be used in court?
It is built for it. Every raw artefact is hashed into a SHA-256 custody chain the moment it lands, evidence packs carry statutory references under BNS and BSA 2023, PII is masked, and access is logged in a DPDP-compliant way. Uncited claims are stripped before a report is written.
Does it hallucinate sources?
Every source returns exactly one of three states: a real result, a checked-and-genuinely-empty result, or unavailable — blocked by a sign-in wall, a bot challenge, a rate limit or a paywall. If information cannot be verified, it is not presented as a fact.
Where is CrypsisX based, and is data kept in India?
CrypsisX Cyber Technologies Private Limited is incorporated in India, with its registered office in Nashik, Maharashtra. Statutory references, PII masking and DPDP-compliant access logs are how the output is constructed, not a localisation layer added afterwards. An air-gapped build for central agencies is on the roadmap.
How do I get access?
Through the contact page or founder@crypsisx.com. Bring a real open matter — that is a better demo than anything we could script. Pricing is shared on request; the engine is never tiered, only case load, seats and support are.
Tell us what you
actually investigate.
We will show you what that looks like on Agent 47.