CLERINT Notebook
CLERINT NOTEBOOK · Standalone nodal analysis

The network is already in the data. Read it out.

A free-standing canvas for link analysis. Import call records, bank statements and travel data, map the columns once, and get a typed entity graph you can pivot, annotate and defend — then fuse several of them into one case.

Start fromCDRFinancialTravelEdge listAny CSV
Reads
CDR
A-party, B-party, duration, cell, tower, IMEI and IMSI — detected, not configured
Types
16 + 17
Node and edge types out of the box, extensible with your own
Detects
18
Named typologies, found before any language model runs
Handles
500k
Rows and edges per notebook, with the limit published up front
Import

It already knows what a call-detail record looks like.

Point Notebook at a CDR dump, a bank statement or a travel manifest. It detects the columns, you confirm the mapping once, and the file becomes a typed graph rather than another spreadsheet.

Call-detail records
Auto-detects A-party and B-party, timestamp, duration, call type, cell ID, tower, LAC, CGI, IMEI and IMSI — building phone, device and location nodes with the edges between them.
Financial
Debit and credit accounts, amount, currency, account holder, card number, merchant and reference.
Travel and generic
Any CSV with a source and target column, plus edge kind, direction, and coordinates as lat/long, a combined geo field or WKT geometry.
File formats
CSV, XLS, XLSX and JSON, multiple files at a time.
Identity schemes
National identity numbers across multiple jurisdictions, plus passport, driving licence, voter ID and residence permit — each validated against its own format.
Transforms
Phone normalisation to E.164, numeric parsing, scientific-notation repair, case folding and whitespace trimming — applied on the way in, once.
Published limits
64 MB per file, 500,000 rows and 500,000 edges, with a warning past 150,000 rows. The limits are served to the browser, so an oversized file is refused before it uploads rather than after.
Ontology

Sixteen node types, seventeen edge types — and your own on top.

The default taxonomy covers most investigations. When it does not, an organisation defines its own entity and relationship types, with its own uploaded icons, and every board resolves against that catalogue.

NodesPersonOrganizationPhoneAccountLocationDeviceVehicleIdentity recordEmailDomainIPCrypto walletDocumentEventAssetCustom
calledtransferred_totraveled_withownscontrolscommunicatesassociatedemploysmember_oflocated_atfunded_byuses_devicehas_accountfamily_ofsame_as
Social network analysis

Measured, not eyeballed.

Who matters in a network is a calculation, not an impression. Each measure below answers a different question, and every result can be reproduced and defended.

Degree
Who talks to the most people.
Betweenness
Who sits on the paths between otherwise separate groups — the broker.
Closeness
Who can reach the whole network fastest.
Influence
Who is important because important people connect to them.
Clustering coefficient
How tightly knit a subject's immediate circle is.
Community detection
The groups the data actually forms, rather than the ones you expected.
PatternsChainTriangleStarCliqueBridgeFanCommunity
Force directedHierarchicalDagre treeCircularConcentricRadialGridColumnsScatter
Detection

Eighteen typologies, found before any model runs.

These are deterministic detectors with published thresholds, not an AI opinion. Each finding carries the node ids behind it, so clicking one highlights the exact records that triggered it — and if the language model is unreachable, the findings are still there.

Structuring
Three or more transactions deliberately sized just under a reporting threshold.
Threshold avoidance
Amounts clustering below a declared limit across accounts.
Smurfing
Many small deposits from many parties converging on one destination.
Layering
Funds in and straight out again through intermediate accounts to break the trail.
Rapid pass-through
Money arriving and leaving within hours, leaving no balance behind.
Funnel account
One account collecting from many and paying out to few.
Dormant reactivation
An account or handset silent for months resuming activity.
One-way broadcast
A number that only ever calls out, never receives — a tasking pattern.
Activity burst
Traffic far above a subject's own established baseline.
Odd hours
Communication concentrated in the hours a subject is normally silent.
Signal calls
Repeated calls under five seconds — contact made without conversation.
Shared contact
Two subjects who should not know each other calling the same third party.
Device sharing
One handset carrying several SIMs.
Handset rotation
One SIM moving between several handsets.
Cross-modal convergence
The same pair appearing in both the call data and the money trail.
Unresolved duplicates
The same person or number entered twice under different spellings.
Fragmented graph
Disconnected islands that suggest a dataset is incomplete.

Thresholds are published rather than hidden: a signal call is a called edge totalling under five seconds with weight three or more; structuring is three or more transactions in the 8,000 to 10,000 band.

Projects

Questions only a fused case can answer.

Attach several notebooks to a project and every real-world entity appears once. Then the cross-dataset analytics become possible — the ones that need the phone records and the bank statements in the same frame.

Temporal coupling
Events in two datasets that keep happening close together in time.
Co-location
Subjects placed at the same location within the same window.
Convergence
Separate threads arriving at the same entity from different directions.
Unified timeline
Calls, transactions and movements on one axis.
Money and comms overlay
The transfer that follows the call, made visible.
Entity resolution
Automatic merge on exact phone, email, wallet, IMEI, IMSI or account number; name similarity goes to a review queue you can confirm, dismiss or reverse.

A missing precondition is reported as such. “No coupled events” and “your financial data has no timestamps” are different answers, and Notebook does not conflate them.

Outputs & custody

The case leaves in one file, analysis included.

Notebook never ships a model key. Point it at the provider your organisation already has under contract; credentials are sealed at rest.

OpenAI
Anthropic
Google Gemini
Azure OpenAI
Groq
Mistral AI
DeepSeek
Cerebras
xAI
OpenRouter
Together AI
Fireworks AI

Platform names and logos are trademarks of their respective owners. Their presence here identifies the sources CLERINT can lawfully collect from, and does not imply any partnership, endorsement or affiliation.

Canvas and table
PNG, JPG, HTML and CSV.
Case archive
A portable .clnb file — the notebook, its nodes and edges, the analysis artifacts, the type definitions with their icons, the raw records, and a manifest of checksums.
Re-import
Archives carry an archive id, a content hash and a lineage id, so re-importing the same file is recognised rather than silently duplicated.
Excluded by design
Model credentials, organisation and user ids, and derived identity keys never travel inside an archive.
Classification
Every notebook carries one, defaulting to UNCLASSIFIED.
Isolation
Sharing never crosses an organisation boundary.
Common questions

The things people ask before a briefing.

Does it read call-detail records?

Yes, and it detects the columns itself — A-party and B-party, timestamp, duration, call type, cell ID, tower, LAC, CGI, IMEI and IMSI. You confirm the mapping once and the file becomes a graph of phones, devices and locations.

What file formats can I import?

CSV, XLS, XLSX and JSON, several files at a time, up to 64 MB and 500,000 rows per file. The limits are published and enforced before upload rather than after.

Do I need a CLERINT subscription to use it?

No. Notebook stands on its own. It reads a CLERINT case when one is connected, but it does not require one.

Can I define my own entity types?

Yes. The sixteen node types and seventeen edge types are a starting point; an organisation can define its own types with its own uploaded icons, and every board resolves against that catalogue.

Can I take a case out of the system?

Yes — as PNG, JPG, HTML or CSV, or as a portable .clnb archive containing the graph, the analysis, the type definitions and the raw records with checksums. Model credentials and identity keys are deliberately excluded.

Bring one dataset.

A single CDR dump or bank statement is enough for a walkthrough — we will import it with you and show you what the graph says about it.