Tips

Here are some tips and best practices for using Gemini Coder effectively.

Context

• Avoid context pollution, for the best performance keep it around 20k tokens.
• Use saved contexts when you often switch between features in a monorepo.
• Include files that demonstrate your team's coding conventions.
• Monitor the token count to stay within reasonable context lengths (~15k).

Chat

• Compare responses across different AI platforms for challenging problems.
• Create task-specific presets (debugging, refactoring, explaining).

Fill-In-the-Middle (FIM)

• Learn when smaller, faster models are sufficient and when you need to resort to priecier, slower ones.
• Bind the FIM completion command to a keyboard shortcut for faster access.
• Use inline documentation comments above cursor.

Refactoring and applying changes

• Review the applied changes before commiting.