ADR-0012: Licensing and project governance

• Status: Proposed (no decision; pyproject.toml deliberately declares no license)
• Date: 2026-10-29

Context

GPUlayer is intended to be open source. The license affects adoption by companies, cloud providers' ability to offer it as a service, contributor expectations, and any future commercial model (for example, a hosted control plane). No commercial model has been decided, and CLAUDE.md asks not to pick one prematurely.

Options

License	Pros	Cons
Apache-2.0	Permissive, explicit patent grant, standard in infra and ML (PyTorch is BSD, Kubernetes is Apache).	Allows proprietary forks and hosted resale without contribution back.
MIT	Simplest and most permissive.	No patent grant; same resale concern.
MPL-2.0	File-level copyleft: modifications to GPUlayer files stay open while embedding stays easy.	Less familiar; some corporate policy friction.
AGPL-3.0	Network copyleft discourages closed hosted forks.	Many companies ban it, which hurts adoption of a developer tool and its workers.

Governance questions to settle alongside:

• Contributor terms: DCO sign-off vs a CLA (a CLA enables relicensing but adds friction).
• Maintainer model, decision process (ADRs), and security disclosure policy (SECURITY.md).
• Trademark policy for the name "GPUlayer".

Decision

None yet. Until decided, the repository must not be published under an implied license.

Consequences

• External contributions should wait for the license and contributor-terms decision.

Follow-up

• Owner decision on commercial intent (hosted offering or not), then pick a license, add LICENSE, the contributor terms and SECURITY.md, and update pyproject.toml.