Generative AI Contribution Policy

This page is intended to be read by human contributors.

Generative AI tools can help you contribute to Grafana Alloy. This page explains what we expect when you use AI assistance.

Core principle

Humans propose changes and remain accountable. AI may assist with implementation.

By submitting a contribution, you vouch for it as your own work. You are expected to understand it, review it, and defend it in discussion, regardless of which tools helped you write the code.

Contribution work has two sides.

• Factual and mechanical work (what changed, implementing code or docs from clear intent, a brief PR summary) is where AI is often well suited.
• Subjective and human work (motivations, trade-offs, reviewer notes, and discussion) stays with you. Use AI where it helps; keep human oversight and interaction where judgment and accountability matter.

Ownership model

You own (write yourself)	AI may help with
Intent and design of the change	Exploring the codebase
Pull request details (why, how, decisions)	Writing or refactoring code and tests
Issue and proposal text you submit	Editing documentation files in the repo
Review replies and ongoing discussion	Brief description of the PR (factual summary)
Code review conclusions	Issue(s) fixed (Fixes #N when known)
Conventional Commit PR title

AI-assisted implementation is welcome when you review and understand the result. AI-mediated conversation is not. Reviewers need confidence that you understand the design choices and trade-offs of what you're proposing.

Acceptable

• Use AI to implement or refactor code and docs that you then review and refine.
• Use AI to draft the PR Brief description and Issue(s) fixed from the actual change.
• Use AI to learn the codebase before contributing or reviewing.
• Use AI privately to clarify your own thinking, then post discussion in your own words.

Not acceptable

• Leave pull request details empty of your own reasoning on non-trivial PRs, or paste AI text there as a stand-in for how/why you built the change.
• Paste AI-generated text into review threads or issue discussion as a stand-in for your analysis.
• Wire an agent to reply to reviewers on your behalf.
• Use AI as a substitute for your judgment when reviewing others' code.
• File automated, bot-driven issues or PRs from tools the Alloy team has not approved.
• Submit PRs or issues that ignore the project templates.

Approved bots (for example dependency bots) are fine when clearly marked as bot-generated.

Disclosure

When AI generates the bulk of the implementation, check "This pull request was substantially generated with AI assistance" in the PR template. Minor autocomplete or small edits do not need disclosure.

An AI-written Brief description and Issue(s) fixed lines are fine. Disclosure does not excuse AI-written pull request details or discussion — those must still be yours.

Licensing and provenance

Alloy is Apache 2.0. Contributions require a signed CLA. The same obligations apply to AI-assisted work: you warrant that you have the right to contribute it, that it is license-compatible, and that any new dependencies follow the dependency guidance.

If you are unsure whether generated code is original or license-compatible, do not submit it.

Alloy tip

LLMs often invent Alloy component names, arguments, and config syntax. Point tools at the Alloy docs and the relevant source in this repo, and validate output against real component definitions and project checks before submitting.

For new components and larger changes, follow the proposal process. AI may help you draft; you own the argument in public discussion.

Enforcement

Maintainers may close or request changes on contributions where pull request details or discussion look unowned or low-effort AI-generated, using the same issue triage process as other contributions. When they do, they will explain why and, where appropriate, offer guidance on how to improve the contribution. Repeated abuse may lead to stricter review or blocked contributions.