What happens to the model if we stop using Akka?

The weights stay in your environment and remain servable. The BSL license on the runtime keeps the right to run, modify and self-host the platform as well.

Can the model be served outside Akka?

The tuned model derives from an open-weight base, so any runtime that serves that base can serve it. Routing, grading and the evidence record are the parts that stay on the platform.

Is our traffic used to train models for anyone else?

Training runs inside your environment on your records, and the model it produces belongs to your system.

Who owns the data that trained it?

Production records stay in the environment that captured them, under the residency and retention rules already applied to them. Sanitizers redact and mask before an event is written.

Does the sovereign deployment change any of this?

Sovereign deployment keeps traffic and data in region with data isolation, networking isolation and local support, and the ownership of the weights is the same in every region.

Guides / Models and routing

Who owns a model trained on your data, and can you take it elsewhere?

Answer

You own the model. The model is tuned from an open-weight base inside your own environment. The weights that result are yours to keep, to serve and to move.

Akka Optimize trains and serves the model on the runtime you already run. The traffic used to train it stays in the environment that produced it, and the model is not held behind a vendor API.

The mechanism

The base license and the training location decide the ownership.

A model tuned from an open-weight base inherits the license of that base. The weights the training produces are an artifact you hold, in the same way the traffic they were trained on is.

Where the training runs decides where the data goes. Akka installs inside your own cloud, datacenter or Kubernetes infrastructure, and inference, training, grading and scoring all run there.

Akka is not a public model-serving marketplace and does not pretrain foundation models. The work Akka Optimize performs is reinforcement learning, fine-tuning and distillation against your production and synthetic records.

What you hold at the end

Each part of the loop is an artifact in your environment.

The weightsThe tuned model, produced from an open-weight base, held in your environment and servable on any runtime that serves that base.

The training recordsYour production and synthetic records, which stayed in the environment that captured them.

The evaluationsThe criteria and the grades that produced the model, which is what lets you retrain it or verify it later.

The runtimeThe Akka runtime is licensed under BSL, so you keep the right to run, modify and self-host it in your own environment.

What this changes

Moving the model elsewhere requires nothing from Akka.

A model tuned inside a vendor service is reachable through that vendor. The accuracy it gained on your traffic is available for the life of the contract that provides access to it.

A model you hold has a life independent of the platform that produced it. The weights, the records and the evaluations are all in your environment, so moving the model to another runtime requires nothing from Akka.

Your investment in specifications carries the same property. A specification describes the system in terms of what it does, so agents stay portable across deployments and across Akka versions.

Questions

Related.

Guide