Public sector and critical

Your agency’s AI.

On your ground, under your control. For government, cities and critical infrastructure, and defense only through cleared partners.

An empty council chamber, its curved oak table lit by a tall window

Where it runs

  • Runs onYour own servers
  • Closed networkTrained on your GPU
  • Air-gappedPlanned
Illustrative. The run modes are in development.
About four weeks for a Test Lab, on your own data
4
Stages, each ending in your decision
3
To roll back; the release you left stays on record
60 s
A site a month, no token bill, proposed
$1,100

What we hear

“This work cannot leave our ground, and we must show who owns every version.”

The work you cannot hand to a foreign cloud. Today it gets generic AI, or none.

  • Residents’ filesCases, permits and records.
  • Network and plant dataEnergy, water and transport.
  • Patient recordsHospital notes, letters and coding.
  • Restricted materialSecurity work, through cleared partners.

The rules

What changes from 2026 to 2028.

  1. Today, 26 October 20262 August 2026Transparency duties apply.
  2. 2 December 2026New prohibitions apply.
  3. 2 August 2027National AI sandboxes are due.
  4. 2 December 2027High-risk duties for Annex III uses, among them critical infrastructure and essential public services.
  5. 2 August 2028High-risk duties for Annex I products.

EU AI Act as amended by Regulation (EU) 2026/1744, OJ 24 July 2026. Eigna is designed to support your compliance work, not to replace it.

The Formula, for you

Your work makes an open model yours.

Base modelAn open model whose licence and origin you can check.
Your workYour cases and files, approved by your experts.
Your agency’s modelTrained by us, owned by you, with a Model Deed.
Base model
An open model whose licence and origin you can check.
Your work
Your cases and files, approved by your experts.
Your agency’s model
Trained by us, owned by you, with a Model Deed.

The model is yours, with a signed Deed.

The Formula

How your people use it

Every release on record.

Each version, its test result and who approved it. Your auditors can read it.

  1. 01ApproveYour experts decide what it learns.
  2. 02GateA new version must beat the current one on your sealed test.
  3. 03RecordEvery release, its result and its approver, kept.
Case expert
Release record
  • Release 8

    Learning from approved cases

    Learning
  • Release 7

    Passed the gate, approved by the owner

    Released
  • Release 6

    A tie: current kept

    Held
  • Release 5

    Rolled back in 60 s, kept on record

    Rolled back
Illustrative record. The Console is in development.

What you will ask

Five objections, answered plainly.

  1. 01

    Data

    This work cannot leave our ground.

    With Eigna Local the data stays on your machines. On a closed network with your own GPU, training stays there too. An air-gapped profile is planned.

  2. 02

    Team

    We cannot build this ourselves.

    You do not need to. We deliver with integrators and, for defense and security, only through cleared partners. Your side owns the task, the data and every release.

  3. 03

    Quality and risk

    We answer for every version.

    A new version goes live only when it beats the current one on your sealed test. Roll back in 60 seconds; the release you left stays on record.

  4. 04

    Cost

    Budgets are fixed a year ahead.

    A fixed running fee per site, no token bill. Proposed: Eigna Local $1,100 a site a month and $3,250 setup per site.

  5. 05

    Oversight and audit

    Oversight will ask who owns it.

    A named owner, a Model Deed and a release record for every version. Designed to support your compliance work, not to replace it.

An Eigna Box on a desk by a window at dusk

Where it runs, for you

On your ground: local, closed or air-gapped.

For the most sensitive work, the data and the model stay on your ground.

Eigna Local
On your machines or an Eigna Box. Data never leaves.
Closed network
Trained on your own GPU, so the data stays on your network.
Air-gapped
No outside connection at all. Planned.

Security and control you can check.

In the design

  • Your keysDesigned
  • Your data trains only your modelDesigned
  • Release record and rollbackDesigned
  • Signed installerDesigned

Planned

  • Audit log of every requestPlanned
  • Air-gapped profilePlanned
  • Refusal training, with a red-team refusal rate at every updatePlanned
  • Certifications, with a published planPlanned

In the design means part of the product design and our release method. Planned means not built yet.

Proof

Published, losses included.

17.8%

Our first training run cut forecast error 17.8% against its base model. It is not yet ahead of a simple last value baseline, and we publish that too.

Base model1.750%
Our first training run1.439%
Last value baselineStill ahead of ours1.392%
Benchmark 001 · forecast error (CRPS), lower is better · 243 held-out days, hourly BTC and ETH, 24 hours ahead · 14 October 2026. An internal diagnostic, not a product claim.

Our first client: a forecasting model for trading. It shares founders with Eigna, said up front. Then a media agency, for image and video, and a web agency, for code.

How it grows

A programme in three stages, each ending in your decision.

  1. About four weeksTest LabOne task, your data. It ends with a report you keep.
  2. ThenPilotOne team, live on your ground. Every release on record.
  3. When you chooseRolloutMore teams, sites and tasks.

Durations are indicative until the first programmes run. We confirm the buying route with your procurement team: a direct Test Lab, an integrator’s framework, your association or a cleared partner.

Start

Start with one task, on your own ground.

About four weeks on your own data. If it does not beat your baseline, you keep the report.