Development assistance

GDevelop AI Agent.

AI assistance inside a visual game engine, focused on making changes to scenes, objects, and event logic in an existing project.

What it does

GDevelop’s current AI documentation describes an agent that reads a project’s objects, behaviors, variables, and events, then works through requests in visible steps. For larger requests, it can present a plan with progress for each step. The older announcement used the “Build for me” terminology; consult the current manual for the interface.

Where it fits

This is development assistance within a game engine. It belongs alongside prompt-led game creation products, but the workflow is different: the project structure remains part of what the creator works with and reviews.

Limits worth understanding

The source explicitly cautions that complex systems need refinement, generated changes require testing, and AI requests consume credits. It also notes that community extensions may not always be supported. Do not interpret the feature as a guarantee of a finished game from one request.

Our assessment status

We have checked both the historical announcement and current manual, not tested the agent. Interface descriptions below come from the manual checked on October 10, 2026.

Control over changes and recovery

Auto edit applies changes directly when enabled. With it disabled, the agent requests confirmation for individual edits. For cloud projects, the manual describes saved restore points before messages and AI requests. Before an assessment, keep a separate project copy and confirm which recovery options apply to that project type.

The manual also distinguishes Medium reasoning from High (Gold or Pro) and Maximum (Pro). Higher reasoning uses more credits; it should not be treated as a guaranteed fix for an unsuccessful generation.

Publishing and plan limits

GDevelop’s manual export documentation describes exporting a browser build to a local folder for hosting elsewhere. Its current plan comparison separately lists AI allowances and automated publishing limits.

This separation matters when evaluating cost: an AI allowance, a hosted publishing service, and a manual export path solve different problems. The availability of a manual path does not establish that every deployment target is one click away.

The manual HTML5 export produces a folder whose entry point is index.html. Upload the complete build, including its resources, rather than only that file. For a proposed portability test, serve the folder over HTTP and inspect missing resources and network requests before publishing it on a separate host.

How to evaluate the workflow

For a first assessment, use a project small enough to inspect. Request one behavior, read the resulting event logic, then change a parameter and test it again. If the result is understandable, the tool may fit an iterative workflow; that remains a project-specific judgment.

We have not run a generation or exported a project. Current checkout costs, credit consumption, and applicable commercial terms should be verified for the intended use.