Add BYOK custom AI provider support for Ask AI

[aaronson2012]

Summary

This PR adds bring-your-own-key support to Ask AI, so users can configure an OpenAI-compatible provider and use whichever model they want instead of being limited to the default GDevelop AI provider.

The feature adds:

• a new AI Providers section in Preferences, with common presets plus custom OpenAI-compatible provider settings
• support for saving, updating, testing, selecting, and deleting provider configurations
• a Custom Model option in the Ask AI model selector when a saved provider is available
• request plumbing so custom provider requests carry the selected provider configuration and do not consume GDevelop AI credits
• local/localhost provider support, keeping local provider keys/configuration on the machine and sending compatible requests directly to the provider
• local Custom Model handling for chat/build flows, including tool-call support and event-generation repair fallbacks where needed

The main goal is BYOK: let people plug in their own provider/API key/model and keep using the Ask AI workflow. The rest of the changes serve that path: preserving configuration across request refreshes, handling provider availability errors, avoiding uploads for local providers, and keeping the UI honest about who may bill the request.

Testing

I ran the focused test set:

npm test -- --runInBand --watchAll=false src/AiGeneration/AiProviderConfigurations.spec.js src/AiGeneration/AiRequestChat/AiConfigurationPresetSelector.spec.js src/MainFrame/Preferences/AiProvidersPreferences.spec.js src/AiGeneration/Utils.spec.js src/Utils/GDevelopServices/Generation.spec.js src/EditorFunctions/EditorFunctions.spec.js

This passed with 6 suites and 101 tests.

I’ve also tested the feature manually, and everything seems to work with no regressions I was able to notice.

[aaronson2012]

AI Preferences Menu

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: fbfb523fd5

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[aaronson2012]

Implemented the Codex review feedback in ef141c3ac:

• network/CORS/DNS failures no longer fall back to local provider configuration storage
• failed local event repair attempts restore the project before retrying
• the provider field now says “Max output tokens” instead of “Context window”
• local Build tool schema now requires object_name for put_2d_instances

Verified with:

npm test -- --runInBand --watchAll=false src/Utils/GDevelopServices/Generation.spec.js src/MainFrame/Preferences/AiProvidersPreferences.spec.js src/EditorFunctions/EditorFunctions.spec.js

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ef141c3ac9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[aaronson2012]

Implemented the remaining Codex review feedback in 079406ee0:

• local add_scene_events now disables content uploads so local providers receive project, extension, and events context inline
• local Custom Model Stop now suspends the in-memory request without calling the backend, and late provider responses no longer overwrite the suspended state
• local restore/fork now creates a trimmed in-memory fork so follow-up context matches the restored project state

Verified with:

npm test -- --runInBand --watchAll=false src/Utils/GDevelopServices/Generation.spec.js
npm test -- --runInBand --watchAll=false src/EditorFunctions/EditorFunctions.spec.js
npx eslint src/AiGeneration/UseGenerateEvents.js src/Utils/GDevelopServices/Generation.js src/Utils/GDevelopServices/Generation.spec.js --max-warnings=0
npm run flow

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 079406ee00

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[aaronson2012]

Implemented the latest Codex review feedback in a0645bc1c:

• strict-JSON tool fallback history now serializes previous tool calls/results as plain text, without tool roles or tool_calls fields
• saving a saved remote provider as localhost now requires a replacement API key
• new local Custom Model requests are stored as working requests before the first provider call resolves, so the UI can select and stop them
• local build chats can now store restore-point project version metadata
• editing a local provider to a remote URL now creates a backend provider configuration, reusing the locally saved key when needed

Verified with:

npm test -- --runInBand --watchAll=false src/Utils/GDevelopServices/Generation.spec.js src/MainFrame/Preferences/AiProvidersPreferences.spec.js src/AiGeneration/Utils.spec.js
npx eslint src/Utils/GDevelopServices/Generation.js src/Utils/GDevelopServices/Generation.spec.js src/MainFrame/Preferences/AiProvidersPreferences.js src/MainFrame/Preferences/AiProvidersPreferences.spec.js src/AiGeneration/AskAiEditorContainer.js src/AiGeneration/AskAiStandAloneForm.js src/AiGeneration/Utils.js --max-warnings=0
npm run flow