Axiris is preparing its AI router: the best engine for each task

6 octobre 2026 · Kris Wilhem

Artificial intelligence models are moving too fast for a company to reasonably build its whole strategy around a single supplier.

One model excels today at image generation. Another is better at editing. A third dominates in video. Some can be run locally. Others are only available in the cloud.

At Axiris, we are therefore preparing a different architecture:

no longer imposing one model on the client, but choosing the best engine for each task.

One platform, several engines

The goal is simple.

The client uses Axiris.

They do not have to manage several subscriptions, several API keys, several interfaces or several credit systems.

Behind the platform, our routing layer will be able to rely, as needed, on:

  • our local models and workflows;
  • private LoRAs and resources;
  • ComfyUI;
  • specialised cloud models;
  • external APIs;
  • routing services such as Comfy Router;
  • future engines still unknown today.

For the client, this complexity stays invisible.

They ask for a result.

Axiris chooses how to produce it.

The best engine in the right place

Not every task asks for the same thing.

An advertising image may require:

  • very high fidelity to the product;
  • perfectly legible text;
  • several reference images;
  • a localised retouch;
  • a video animation;
  • a final 4K upscale.

There is not necessarily a single ideal model for the whole process.

Our goal is therefore to be able to use several engines within the same production chain.

For example:

```text CLIENT BRIEF │ ▼ NOVA / AXIRIS │ ├── needs analysis ├── business context ├── confidentiality ├── expected quality ├── cost └── deadline │ ▼ AXIRIS ROUTING │ ├── local model ├── ComfyUI workflow ├── cloud API ├── specialised image engine ├── specialised video engine └── upscale / post-production ```

Technology thus becomes an interchangeable resource in the service of the project.

Sovereignty where it really matters

At Axiris, opening up to the cloud does not mean giving up local.

Quite the opposite.

We want to keep under our control what constitutes the system's own value:

  • business knowledge;
  • the Brain and the knowledge base;
  • sensitive data;
  • proprietary workflows;
  • LoRAs;
  • private models;
  • production rules;
  • the resources specific to each client.

The cloud steps in when it brings a real advantage in quality, speed or capacity.

Sovereignty therefore does not necessarily mean:

"everything must run locally"

but rather:

"we keep control of what matters, and we freely choose where to run the rest."

A logic designed for Swiss companies

For a company, the difficulty is not only to access the most recent models.

The difficulty is to integrate them cleanly into a production environment.

That involves:

  • confidentiality;
  • costs;
  • access rights;
  • traceability;
  • stability;
  • infrastructure;
  • support;
  • the constant evolution of models.

Our goal is that the client does not have to manage this complexity themselves.

Axiris must be able to offer them a single AI production layer, suited to their business and to their level of confidentiality.

Some tasks will be able to stay entirely on local or Swiss infrastructure.

Others will be able to go to a cloud engine when it brings a real advantage.

A single bill

This approach also changes the way AI is billed.

The client should not have to buy separately:

  • credits from several suppliers;
  • subscriptions to different services;
  • several GPU infrastructures.

We are therefore working on a logic of Axiris credits.

Each task consumes different resources, but billing stays centralised in a single platform.

Behind a generation, Axiris can measure:

  • the supplier's cost;
  • local GPU time;
  • storage;
  • processing;
  • the resources needed for production.

The client buys production capacity.

Not an API call.

The market is starting to move the same way

The recent launch of Comfy Router is an interesting signal.

Comfy now offers a single interface giving access to several models and several suppliers.

This approach confirms an evolution we consider essential: value is gradually shifting from the model itself to the layer able to choose and orchestrate the right models.

At Axiris, Comfy Router may become one of the services our infrastructure can use.

But it will not replace our logic.

Our router must also know how to choose between:

  • cloud;
  • local;
  • proprietary workflow;
  • private model;
  • client resource.

What the client really buys

In time, our goal is that the client can simply say:

"I want this campaign, with these constraints, these references and this level of confidentiality."

They do not have to know whether a step used a local model, an external API, a ComfyUI workflow or several engines in succession.

Making that choice is our job.

The value of Axiris is therefore not in owning every model.

The value is in knowing:

which model to use, when, why, with which data and at what cost.

That is the layer we are building today.