Agidel · Secure AI for finance

Gnostic.

Builds the knowledge your organization runs on — inside your organization.

Documents, messages, the knowledge in people's heads — Gnostic connects it all into one living knowledge graph that empowers your teams and AI agents. The result: lower operational costs, less risk, and higher productivity.

Deployment Our cloud, your cloud, or on-premise. Full air-gap available.
Economics Pay per document ingested — never per question asked.
Domain A knowledge model built around your business — not a generic template.
Who it's for

Built for document- and knowledge-heavy operations.

01 Commodity trading houses

Hundreds of live shipments, each trailing a stack of paperwork. Know what's complete on every deal at a glance.

02 Trade finance banks & lenders

Exposure per counterparty and documentation gaps across every active deal — surfaced before the audit finds them.

03 Heads of ops & compliance

Drowning in manual document checks, answerable to a regulator. Every answer has to carry its evidence.

04 CTOs at financial institutions

Board pressure to "do AI" — and a policy that forbids sending data to public models. Both, satisfied at once.

The problem

The answer is always somewhere in the documents. Finding it is somebody's entire job.

— what we hear in every first conversation with a document-heavy team

01

The same fields, read again and again

Specialists reopen the same files to pull the same facts, operation after operation, deal after deal. The knowledge evaporates the moment the file closes.

02

Connections rebuilt by hand

Same counterparty, same deal — scattered across drives, email, and systems. Every link is rebuilt from scratch, every time.

03

And public AI is off the table

Your data would leave your jurisdiction, with no audit trail and costs that climb per question. Compliance is right to say no.

What you get

Not a better search box. The knowledge itself — connected.

A living knowledge graph

Every party, contract, shipment and payment, extracted and linked — and kept current as new files arrive.

Answers with receipts

Every answer traces to its source documents. Structured facts, not generated guesses.

A view for every role

Compliance sees gaps. Operations sees deadlines. Financing sees exposure. One graph, many views.

Gaps flagged before deadlines

Missing certificates, approaching payment dates — surfaced automatically, not at audit.

How it works

Ontology first. Then the pipeline.

It starts from a pre-built domain ontology — your document types, entities and relationships, modelled before a single file is read. Four stages run on top of it.

  1. 01

    Watch

    Drives, email, systems — every new or changed document is noticed the moment it lands.

  2. 02

    Extract

    Each document is classified and its facts pulled into structure — against your domain, not a generic template.

  3. 03

    Connect

    Facts become a graph: parties, contracts, shipments and payments, linked. This is the knowledge — built once, kept live.

  4. 04

    Answer

    Questions read the graph directly. Nothing is reprocessed, so the next answer is effectively free.

To be clear

What Gnostic isn't.

A chatbot

Answers come from a structured graph, queried directly. Nothing to hallucinate.

RAG over a document dump

We model your domain first. Extraction follows the ontology — not keyword similarity.

A public AI subscription

Nothing leaves your perimeter. Local models supported end-to-end, air-gap included.

A document manager

We don't move or replace your files. We connect what's inside them, wherever they live.

Case study · in development

Commodity trade finance, first.

We're building Gnostic with a launch partner — a major commodity trading house, in one of the most document-heavy corners of finance. The domain came first: every document type and relationship modelled before a line of pipeline code.

Why start here? Because if it works in one of the most document-heavy corners of business, it works anywhere knowledge is buried in documents and people's heads.

gnostic — trade-finance graph

Which trades with Meridian SA are still missing a certificate of origin?

Result3 trades. Oldest open: 22 days.
Trades #2841, #2867, #2903 — sources attached.

Illustrative queries from the trade-finance pilot. Every answer is traceable to its source documents.

How an engagement runs

Model first. Then pilot. Then production.

Phase 01 Domain modelling

We map the territory with your specialists — document types, entities, relationships. The ontology is the output, and it's yours to keep.

Phase 02 Pilot on real documents

A real corpus from your archive, ingested into your environment. Your team asks real questions and checks the sources.

Phase 03 Production deployment

Live document flow connected, role views configured, your IT in control. Cloud, private, or fully air-gapped.

During modelling we need a few hours a week from two or three of your specialists. We do the rest.

Agidel services

The team behind Gnostic also builds to order.

Agidel implements AI for financial institutions — same discipline as the product: model the domain first, stay inside your perimeter, keep every output auditable.

Assess

Feasibility & architecture

Where AI genuinely pays off — and where it doesn't. Honest scoping and an architecture that fits your compliance posture.

Build

Custom AI solutions

Document processing, knowledge extraction and workflow automation — on your data, your infrastructure.

Operate

Secure deployment & operations

Private-cloud and on-premise AI, air-gap included. We run it, or hand it to your team.

A quick fit check

Where this works best.

And sometimes we're not the fit. Want a general-purpose chatbot for everyone? A copilot licence is cheaper. Data already in one clean database? Just query it. Either way, we'll tell you on the first call — and point you to who can.

Questions we get

Asked before every engagement.

How is Gnostic deployed, and where does our data live?

Most teams run Gnostic as a managed service in our secure cloud — the fastest way to get value, with Swiss data centres available. If your compliance needs more, we can deploy inside your own cloud account, on your servers, or in a fully air-gapped environment where nothing leaves your perimeter. The right model for you is settled in the architecture phase.

Which AI models does it use? Can we run local ones?

The pipeline is model-agnostic. In connected environments it can use the frontier model of your choice under your agreements; in private and air-gapped deployments it runs on local models end-to-end. Model choice is part of the architecture phase.

How do we know the answers are correct?

Every answer is a query against structured data, and every extracted field links back to the document it came from. Your team verifies against sources during the pilot — that verification step is built into the engagement, not skipped past.

We're not in commodity trade finance. Is this still for us?

The approach — ontology first, then extraction, then the graph — is domain-independent. Trade finance is where the pre-built ontology exists today; for adjacent domains like project finance, syndicated lending, or structured products, modelling your domain is the first phase of the engagement.

What does it cost?

Pricing follows document volume ingested, not questions asked — so the cost is fixed and plannable, and no one has to ration their curiosity. Design partners get preferential terms; exact numbers depend on corpus size and deployment, which we'll size up on a first call.

Do you implement, or only advise?

Both, but always toward something running. Agidel is an engineering team — backgrounds in finance, production AI systems, planet-scale infrastructure, and AI security. Advisory exists to scope the build, not to replace it.

Get started

See Gnostic on your own documents.

Start with a focused pilot — a slice of your real corpus, ingested inside your environment, your team asking real questions and checking the sources. A few hours a week from your specialists is all it takes to begin, and early teams get preferential terms and an ontology tuned to their business.

A founder reads every message. You'll hear back within 48 hours.