avante.nvim/cursor-planning-mode.md

Cursor planning mode

Because avante.nvim has always used Aider’s method for planning applying, but its prompts are very picky with models and require ones like claude-3.5-sonnet or gpt-4o to work properly.

Therefore, I have adopted Cursor’s method to implement planning applying. For details on the implementation, please refer to: 🚀 Introducing Fast Apply - Replicate Cursor's Instant Apply model

So you need to first run the FastApply model mentioned above:

bash~~ ollama pull hf.co/Kortix/FastApply-7B-v1.0_GGUF:Q4_K_M ~~

An interesting fact is that I found the FastApply model mentioned above doesn't work well. First, it's too slow, and second, it's not accurate for processing long code file. It often includes // ... existing code ... comments in the generated final code, resulting in incorrect code generation.

The best model I found for applying is qwen-2.5-coder-32b on Groq, it's both fast and accurate, it's perfect!

Then enable it in avante.nvim:

{
    --- ... existing configurations
    cursor_applying_provider = 'groq', -- use groq for applying
    behaviour = {
        --- ... existing behaviours
        enable_cursor_planning_mode = true, -- enable cursor planning mode!
    },
    vendors = {
        --- ... existing vendors
        groq = { -- define groq provider
            __inherited_from = 'openai',
            api_key_name = 'GROQ_API_KEY',
            endpoint = 'https://api.groq.com/openai/v1/',
            model = 'qwen-2.5-coder-32b',
            max_tokens = 8192, -- remember to increase this value, otherwise it will stop generating halfway
        },
    },
    --- ... existing configurations
}