Is it a fit?Book a call

Owned AI

AI of your own.

Your own model. It learns your work and runs where you decide.

Book a callSee if it’s a fit
  • YoursThe model, the data and the experts belong to you.
  • LearnsA new version goes live only when it beats the current one.
  • ProvenA metric, a baseline and a date. Losses included.

The shift

Every company rents the same AI.

The most important technology of our time is rented by the token. Every company gets the same model as its competitors, and its work trains the vendor’s next one. So the most valuable work gets no AI at all.

Owned AI
ComparedRented AILocal AIOwned AI
Whose modelThe vendor’sOne model for everyoneWhose modelYours, with a signed Deed
Learns your workFor the vendorNeverLearns your workKeeps learning, released only when better
Where it runsIn someone else’s buildingOn your machineWhere it runsWhere you decide, local or cloud
The billPer seat or token, rising with useYour hardware, your teamThe billA fixed running cost, no token bill

Local is where AI runs. Owned is whose it is.

Yours

Your work makes a model of its own.

An open base model, plus the work your people approve, equals a model that belongs to you.

Ownership is a document, not an adjective.

Every model comes with a Model Deed: the owner, the version, the date and the fingerprint of the weights, signed. Each release that passes the gate adds a line. You can take it with you.

The Model Deed

Model Deed

No. 3302

Owner
Keller
Model
Keller TLM
Release
12
Date
22 October 2026
Fingerprint
9f2c 41ab 77e0 c3d1
Signed
Keller and Eigna

Illustrative data. The live version only changes when a release beats it.

Learns

Keeps learning. Released only when better.

Your people approve the work. A new version is trained on it, tested on a sealed set it has never seen, and released only when it beats the one you use today.

  1. 01UseYour team works with its expert every day.
  2. 02ApproveNothing learns until a person approves it.
  3. 03TrainApproved work trains a new candidate.
  4. 04Test at the gateOn a sealed set, against the live version.
  5. 05Release or holdBetter goes live. Not better is held.
Keller TLM
Release record

  • Release 13

    Learning from approved work

    Learning

  • Release 12

    22 October 2026

    Released

  • Release 11

    8 October 2026

    Released

  • Release 10

    24 September 2026

    Released

  • Release 9

    10 September 2026

    Held

Release gate

Release 13 against release 12, sealed test set

Passed
Live, release 12100.0
Candidate, release 1398.6

Forecast error, indexed to the live version. Lower is better.

to roll back to the previous release. That is the design target.

60 s

How it learns

Proven

We publish our results. Including the ones we lose.

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 · run on 14 October 2026.
Case 001: held at the gateTests and benchmarksbenchmarks.eigna.ai

Where it runs

Local or cloud. On your hardware or ours.

We deliver it, set it up and run it. No ML team to hire. No token bill: a cost you can plan.

Compare the three
An Eigna Box against the wall of a quiet officeEigna LocalOn your servers or an Eigna Box on your site.Data never leaves.A glass office lit up inside a dark factory hallEigna HybridSmall experts on your machine, the main expert on a GPU in your region.Data stays in the region you choose.A row of server cabinets in a long, dim server roomEigna CloudA dedicated GPU in the region you choose, run by us.Data stays in the region you choose.
The Eigna Box family: Personal, Home, Assistant, Business and Enterprise

Eigna Box

One base. Five homes for the model.

From a box in the hand to a rack in the data centre: the same model inside, your name on the plate.

Meet the Eigna Box

Where people use it

Wherever the work already happens.

People ask it in the apps they use every day. Teams share one model. Agents run whole tasks with it, on their own and within your rules. Every answer names its sources and the release that gave it.

Hands on a laptop, approving an answer in the Eigna app

Who it is for

For everyone. Above all, for data you cannot give away.

Any company, team or person can own its AI. It matters most where the data has to stay inside: client files, patient records, plans, code and anything under rules.

EnterpriseYour company’s own AI.Small and mid-sizeYour team’s know-how, in a model you own.StartupsOwn the model your product runs on.Public sectorOn your ground, under your control.

What comes next

Every company, then every home.

  1. A row of server cabinets in a long, dim server room

    Now

    Companies and their teams

    The web app with an assistant for each person, on your own servers, in the data centre or in the cloud.

  2. An Eigna Box against the wall of a quiet office

    Next

    Every company size

    Eigna Cloud and Hybrid, the Eigna Box Business in the office, a simpler web app and the mobile app.

  3. Eigna Box Home on a sideboard in a living room

    Then

    People and homes

    The mobile app and the smaller boxes, Personal, Home and Assistant: an assistant that learned your way of working.

People

Engineers first. Evidence by habit.

The team
  • Omar Garcia

    Omar Garcia

    Co-founder and CEO

    Thirteen years of production systems where mistakes cost money. Leads the lab.

  • Andrés Soltermann

    Andrés Soltermann

    Co-founder and CBO

    Owns every route to market: own sales, partners, embedded, ventures.

  • James Sherborne

    James Sherborne

    Co-founder and CFO

    From banking to blockchain infrastructure. Owns the plan, the round and the compute.

  • Roman Fritschi

    Roman Fritschi

    Co-founder and CTO

    Built conversational AI before the LLM era. Turns what we build into products.

From Insights

All insights
JournalBenchmark 001, losses includedOur 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.
JournalHow a release earns its placeA new version goes live only when it beats the one you use today.JournalWhy there is no token billRented AI gets cheaper per token and more expensive per month.

Start

Bring one task. See the proof first.

In about four weeks, on your own data in a sealed setup, we show whether an Eigna expert beats your baseline. If it does not, you keep the report.

  1. 01See if it’s a fitTwo minutes. An honest answer, no email needed.
  2. 02Book a callThirty minutes with a founder.
  3. 03Request a Test LabFour weeks, one task, a report you keep.