it depends on which ai model.. qwen2.5-coder:7b works but 32b is better but slower, check aider benchmarks to have an idea