CobbleBrain - AI Dialogue System for Cobblemon

Major Changes

Complete restructuring of the AI: Now it is possible to use not only Google’s Gemma models for free, but also any provider that uses the same schema of the OpenAI API call. This includes OpenRouter and (obviously) OpenAI itself. Local models also have initial support, but are still in a testing/unstable phase — use at your own risk.

Improved response time
Bug fix that caused delays in responses.

Optimized default settings
Internal adjustments for better performance and stability.

New config values in cobbblebrain.json
Removed visibleAiWarnings → But don't worry, the player receives notifications in chat on whether the requests were not successful.

debugLogging → Records information whenever a prompt is sent to the model, including possible errors. The logs are stored locally in cobblebrain-ai/logs.

apiBaseUrl → Defines the API endpoint that will be used to access the AI model.

aiModel → Specifies which AI model will be used to generate responses.

temperature → Controls the level of creativity and variation of the responses.

aiProvider → Indicates the provider within OpenRouter (ignored in other services).

reasoningEffort → Adjusts the reasoning effort of the model, from simple to advanced. (Ignored on Google API)

ShowFriendship → Allows the player to view the amount of friendship gained or lost in interactions.

lowTokenMode → Removes from the prompt information such as special blocks, mobs, and other entities so that the AI ​​needs fewer tokens to generate the response and can have more focus on the dialogue.

Minor Changes

Summarized error messages to make understanding easier.

Removal of "Pokémon are processing what you said" warning → making it possible to send messages without interruptions.

Fixing various bugs → providing greater overall stability to the mod.