Why BranchPy's core analysis does not need AI

BranchPy's core analysis does not need an AI model. It reads your project and applies checks to the structure and references it can understand. You do not need to connect an AI provider to use those checks.

That makes it useful whether you write every line yourself, work with other developers, or use AI to help produce code. The question is the same: does the resulting project contain problems worth investigating?

Why the checks do not depend on AI

A reference to a missing resource is a concrete thing to investigate. So is a transition pointing to a destination that cannot be resolved. BranchPy can inspect supported project formats and report these relationships without asking a language model to interpret a prompt.

The core analysis is deterministic: with the same project inputs, settings and analysis version, its checks are repeatable. That gives you a stable basis for reviewing changes. If a result changes after an edit, you can investigate what changed in the project rather than wonder whether a model simply phrased its answer differently.

Deterministic does not mean infallible. Dynamic references, custom loading patterns and unsupported behavior can limit what static analysis sees. Some results need your judgment. BranchPy's job is to make the evidence easier to inspect, not to replace your understanding of the game.

A QC pass for code written with AI

Suppose an AI coding assistant helps you add a Godot menu. The script looks plausible, but it refers to a scene that has a different name in your project. Or it helps extend a Ren'Py route and introduces an image reference that has no matching asset.

BranchPy does not need to know who wrote that code. It can check the supported structure and references in the files you actually have. This gives you a review step that does not depend on the coding assistant's own assurance that its work is finished.

A practical workflow is:

1. Keep a baseline. Save or commit your current project before accepting a change.
2. Review the change. Check what the human or AI assistant added, removed or renamed.
3. Run the relevant BranchPy tools. Review the results for the areas affected, within the support available for your engine.
4. Inspect the evidence. Follow the reported references and source locations; resolve uncertainties before treating them as confirmed bugs.
5. Test in the game engine. Exercise the changed behavior, make any corrections yourself, and run the checks again.

This is a quality-control pass, not an AI-content detector or a certificate that the code is correct. It cannot prove that a story choice feels right, that every runtime state works, or that every possible defect has been found. Code review and playtesting still matter.

Checking code does not mean changing it

BranchPy's core analysis does not modify your source code. Its optional AI explanation and review helpers do not edit that code either. You review the results and decide what to change in your editor. This review workflow does not apply fixes automatically.

Read-only analysis refers to your source files. BranchPy can save reports and analysis data; saving a report is different from rewriting a script. Keeping that boundary clear lets you inspect a project without handing over control of its code.

Why offer optional AI assistance, then?

Finding an issue and understanding what to do about it are different tasks. Optional AI helpers, such as warning explanations and code review, can add context, help interpret a result and suggest questions or next steps to investigate. Availability depends on the product surface, provider configuration and feature access; they are not a prerequisite for core analysis. Their advice is separate from the deterministic results and should be reviewed before you act on it. This assistance is advisory, not an instruction for BranchPy to edit your code.

If you choose a connected AI feature, review what it sends to your chosen provider. Local core analysis and an optional provider request have different data flows. See our privacy information and feature availability before enabling it.