Open by Design. Private by Principle.

Our public / private boundary

Matereal.ai maintains a strict separation between what is open and what is private. Open industry knowledge and private customer advantage are kept separate.

Open models are reference models, public-data models, synthetic-data models, research implementations, or Matereal-owned base models. Customer-specific fine-tuning remains private.

What we publish

We publish open reference models, synthetic datasets, research implementations, process explainers, and benchmark tasks.

Everything published is meant to be non-confidential: reference intelligence for learning, benchmarking, and collaboration, never customer secrets.

What we never publish without permission

• We do not publish customer data without written permission.
• We do not publish plant-specific models without permission.
• We do not publish customer performance results without permission.
• We do not expose process recipes, proprietary features, plant operating practices, or plant-specific benchmarks.

Customer data policy

Customer data remains customer-controlled. Customer data is not used to cross-sell, expose, benchmark, or train public models without explicit approval.

Data provided for a private engagement is isolated to that engagement.

Model fine-tuning policy

Customer-specific fine-tuning remains private. Private customer deployments can create customer-specific IP.

A model fine-tuned on customer data is never published, benchmarked publicly, or reused for other customers without explicit written approval.

Public model policy

Public models are reference implementations only. They are trained on synthetic, public, or Matereal-owned data.

Every public model card states its type and carries the reference-implementation disclaimer: industrial use requires validation on plant-specific data.

Benchmark policy

Benchmarks use synthetic, public, or sample datasets. We do not publish plant-specific benchmarks or customer performance results.

Leaderboard scores are illustrative of method comparison on open splits, not claims of industrial performance.

Customer IP policy

Private customer deployments can create customer-owned IP. Customer recipes, operating practices, and results belong to the customer.

Open industry knowledge and private customer advantage are kept separate.

NDA policy

Private engagements operate under NDA. NDA-protected deployment details are never disclosed, referenced, or implied in public materials.

We do not reference customer projects publicly.

Data isolation principles

Customer data is isolated per engagement and not commingled across customers.

Private deployments can run on-prem or in a private cloud, keeping data within the customer's environment.

Open-source contribution policy

Open contributions such as reference models, synthetic datasets, research, and benchmark tasks are welcomed under open licenses.

Contributors are asked to submit only non-confidential material, with no customer data or plant-specific know-how.

Private collaboration policy

Private collaboration covers data readiness reviews, private fine-tuning, plant-specific development, secure deployment, and joint research without public disclosure.

Outputs of private collaboration are governed by agreement and remain private unless explicitly approved for publication.