Which AI connection suits your GameMaker project folder
PixelTread works with the model you connect: a cloud AI provider through a standard API endpoint, a model running on your PC, or the built-in model. Each choice has different consequences for privacy, and you can change it later.
Three ways to connect a model
A cloud AI provider
Connect a cloud AI provider through a standard API endpoint when you want the processing power of a large model for refactors or tricky GML logic. You enter the endpoint and your key in the app, and the key stays on your PC. The privacy consequence is plain: your prompt and the file text the agent reads are sent to that provider. A cloud provider may charge you separately, so check its terms before a long session.
A model on your PC
A model running locally on your PC suits you when scripts and prompts must stay on the machine. You point PixelTread at the local endpoint, and with a local model nothing leaves your PC for the model step. The cost is hardware: a larger local model needs more of your own memory and graphics power. Smaller local models handle short edits well, while long refactors may go better with a bigger model.
The built-in model
The built-in model is ready to use without your own endpoint, so it is the shortest path from installing the app to your first request. Pick it when you want to try the workflow on a small script, such as a movement function, before you decide on a provider. You can switch to a cloud provider or a local model at any time from the same settings. Nothing else changes in how the cards, diffs and approval modes behave.
Check the connection with Test
After you enter an endpoint, press the Test button. It checks that the endpoint answers, shows the response time and shows the context window of the model. The context window matters for GameMaker work because a long script, a build log and your request must all fit into it. If Test reports a small window, keep your requests short and point the agent at one file at a time. If the test fails, correct the address or the key and run it again before you start a chat.
Match the model to the task
Use a bigger model for refactors and complex logic, for example restructuring several scripts that share a movement function. Use a smaller model for small edits such as renaming a variable, adding a comment or rounding a position. You do not have to restart the chat to get both: run_on_model sends one hard step to a bigger model while the rest of the session stays on the smaller one. Try the smaller model first, and move a step up only when the answer shows it needs more.
Quality varies between models, so treat every result as a proposal. Review the diff in the card, and use Ask first mode so that nothing that changes a file happens before you agree. Build and test in the GameMaker editor after each accepted change. A good habit is to accept one edit at a time.
