Cygnus

Private
Enterprise AI

An introduction to the platform — what it is, how it is built, and what you decide.

Before we start

What we will cover.

One

What Cygnus is — and what it is not

Two

The five layers underneath the chat window

Three

The product, working

Four

The decisions that stay with your institution

In one sentence

Cygnus is a private, sovereign AI
environment for an institution.

Your people use AI the way they already know how. The institution keeps control of the data, the permissions, the rules, the models and the cost.

Your people

Accessible only through corporate mailboxes.

Your knowledge

Connected to the systems you already use.

Your rules

Permissions, limits, and a full record.

Your infrastructure

Your own servers, or a private space built for you.

To avoid confusion

What Cygnus is not.

Not a chatbot

Chat is one application sitting on top. The platform is everything underneath it.

Not a search engine

It does not build an index of documents. It builds a map of the things your institution works with.

Not an AI model

Models plug into it and can be exchanged. Cygnus is what surrounds them.

Not a system you move to

Nothing is migrated. It sits underneath what you already run.

The visible part

What your people see.

chatapplication
.
cbainstitution
.
amdomain

Deployed on a subdomain of your own domain — your name, your address, your certificate. A familiar chat window and user experience. Sign-in through corporate mailboxes. A choice of model and how deeply to search. Nothing to learn.

This is the whole of what an employee experiences.
It is a small part of what Cygnus is.
The rest of it
Five layers.
Your people see one.
Layer 2 · Agents

Seven decisions, made by you.

Agent
A governed reader of your knowledge
Seven things you set, explicitly
Knowledge it may read
one or more knowledge bases — and nothing else
The topics it will answer on
the frame — anything outside it is declined
The model it runs on
cloud, or private and inside your building
Instructions
how it should behave, in your words
Tools it may call
quick lookup, or a deep search
Grounding mode
strict — answer only from knowledge, or say so
Usage cap
how much it is allowed to spend
Everything an agent may do is set in advance — including the topics it will answer on, and whether it may answer from anything other than your own knowledge.
Layer 2 · Agents

The model is a slot.

Everything about the agent stays the same
Knowledge·Permissions·Instructions·Tools·Audit·Limits
Only the model changes
Anthropic
Frontier · cloud-hosted
Most capable. Highest cost per question.
Gemini
Frontier · cloud-hosted
Most capable. Highest cost per question.
OpenAI
Frontier · cloud-hosted
Most capable. Highest cost per question.
DeepSeek
Open-weight · runs in your building
Focused answers. A fraction of the cost.
Qwen
Open-weight · runs in your building
Focused answers. A fraction of the cost.
The same answers you can trust. Only the capability and the cost change.
Take one model out, put another in. Nothing else about the agent changes.
Layer 3 · Security

The same question, two answers.

One identical question, asked by two employees
“What did the last supervisory decision on this institution require?”
Head of Supervision
Supervision division
All staffSupervisionLegal
Supervisory decisions
Legal & regulation
Statistics
Board papers
A complete answer, citing the decision, the legal act it rests on, and the board paper behind it.
Analyst
Statistics division
All staffStatistics
Supervisory decisions
LOCKED
Legal & regulation
Statistics
Board papers
LOCKED
“I don't have access to that.” The supervisory record is never retrieved, so it can never leak into the wording.
The AI is only given what the person may see. It cannot mention what it never received.
Two people ask the same thing. The AI reads only what each of them is allowed to read, so the answers differ.
Layer 4 · Knowledge

Connect almost anything.

Where knowledge comes from
Public websites
read and re-read on a schedule
Any system, through its API
a REST connection
Files uploaded directly
a document at a time
Whole folders and shares
a directory at a time
Databases and registries
structured records
Mail, intranet and wikis
everyday working material
If it can be read, it can be connected.
Nothing is migrated.
Your ontology
one per knowledge base — the map of
the things and relationships
your institution actually works with
Governed knowledge
Regulationentity
Decisionentity
Institutionentity
Personentity
Indicatorentity
Publicationentity
… and every relationship
between them.
Websites, APIs, uploaded files, whole folders, databases, mail. If it can be read, it can be connected — and it stays where it is.
Layer 4 · Knowledge

It keeps itself up to date.

01
Read
parse and normalise the source document
02
Identify
find the things your ontology defines
03
Extract
resolve attributes and relationships
04
Commit
write the facts and links into the graph
It never stops
Sources are re-read on the schedule you set. New documents enter the graph on their own. Nobody maintains the knowledge base.
Why this matters for privacy
Every stage is built to run on open-weight models. The entire pipeline — the part that reads every document you own — can run inside your own building.
In one live deployment
14,113
documents
0
people maintaining it
Nobody is given the job of keeping it current. And the part that reads all your documents can run inside your own building.
Layer 4 · Knowledge

A map of what the institution knows.

EconomicIndicatorPublicationBoardDecisionOrganizationFinancialInstitutionStatisticalSeries
Entity types
EconomicIndicator
Publication
BoardDecision
Organization
FinancialInstitution
StatisticalSeries
+ 14 more
26,000
entities
47,000
relationships
And it does not only find
Analyse
summarise a body of decisions, compare documents, show what changed
Reason
connect facts across sources to answer what nobody wrote down
Act
hand routine work to an agent that follows the same rules
Not a search box — a map of what your institution works with. And agents do not only search it.
Where it runs

Public cloud, private cloud,
or your own infrastructure.

Public cloud
Azure, AWS and others

Managed for you

The fastest way to start. No equipment, and nothing to prepare.

Private cloud
Focused on Firebird

An isolated environment

A separate network of your own, joined to yours. Not shared with anyone.

Your own infrastructure
Your servers, your data centre

Inside your building

Everything, including the models, on machines you own and control.

Who does what

Three companies, three roles.

Cygnus
The platform
The product, its engineering and its direction.
VOLO
Integration partner
Connects it to your systems, configures it, and supports it.
Firebird
The infrastructure
Computing power, hosting, and private environments.
The software, the people who fit it to your institution, and the place it runs.
The decisions that stay with you

Five things only your institution can decide.

  • Which knowledge each group of people may reach — and which stays closed, file by file.
  • Which models you allow, and whether they run in the cloud or inside your own building.
  • How much may be spent — weekly and yearly, per team and per person.
  • Which rules the application enforces, and what it tells someone when it stops them.
  • Where the whole thing runs.
Cygnus
The chat window is the part everyone sees.

What makes it safe to use is the four layers underneath it — and the decisions that stay with you.
TheCygnus.ai

Platform: Cygnus  ·  Integration: VOLO  ·  Infrastructure: Firebird

What your people see
A chat window. Nothing to learn.
What makes it safe
Four layers your people never see.
← → navigate  ·  O overview  ·  D demo fallback  ·  F fullscreen
No recording found
Save a screen recording as cygnus-demo.mp4 beside this file to arm the fallback. Press D or Esc to close.

Jump to slide