How Matrix inference works

Matrix’s primary card-generation models were trained and are served on owner-operated consumer workstations. That is the inference work described by the homepage commitments.

Primary inference

• Core evidence card generation runs on owner-operated consumer hardware.
• Matrix does not operate or rent an enterprise data center for that primary model inference.
• The owner-operated inference hardware does not use an external data-center water cooling system.
• Your host agent may itself be local or cloud-hosted; Matrix is compatible with both.

Supporting cloud services

A production internet service still requires supporting infrastructure. Matrix uses third-party cloud services for account authentication, storage, website and network delivery, payments, transactional email, and occasional limited supporting inference. These services are not the primary Matrix card-generation model.

Evidence and updates

Matrix will revise this explanation if its primary inference architecture changes. Questions about these commitments can be sent to support@matrixdebate.com.