Rate limited... but by whom?

[npearson72]

I'm freshly new to this, so please bare with me.

My original provider that I had setup was copilot. Worked really well, until I asked it to fix circular dependency linting errors. It seemed to get stuck in a loop and apparently used up my limit for the month.

I also use gemini pro (not in neovim yet, just via Chrome). So I decided to try that out.

But the message I'm getting from avante is that I'm rate limited. I check with gemini in Chrome, and it said I am not. Was clearly able to respond to questions.

I then clobbered all my neovimi cache (~/local/state, ~/local/share, etc..) and re-installed all my plugins. Still having the same issue. Here is my config:

return {
  {
    'yetone/avante.nvim',
    build = 'make',
    event = 'VeryLazy',
    lazy = false,
    version = false, -- Never set this value to '*'! Never!
    opts = {
      file_selector = {
        provider = 'telescope'
      },
      hints = { enabled = false },
      input = {
        provider = 'telescrop',
        provider_opts = {
          title = 'Avante Input',
          icon = ' ',
        },
      },

      provider = 'gemini',
      providers = {
        -- copilot = {
        --   model = 'copilot/gpt-5',
        -- },
        gemini = {
          api_key_name = "AVANTE_GEMINI_API_KEY",
          model = 'gemini-1.5-pro',
          temperature = 0,
          max_tokens = 4096,
        }
      },
    },
    dependencies = {
      'nvim-lua/plenary.nvim',
      'MunifTanjim/nui.nvim',
      --- The below dependencies are optional,
      'nvim-telescope/telescope.nvim',
      'hrsh7th/nvim-cmp',
      'nvim-tree/nvim-web-devicons',
    },
  }
}

I'm sure it's something super obvious to you guys. I just don't see it.

[edmundhighcock]

Not an expert here, but since you are providing your API key then I guess Gemini will know who you are and will remember that you are rate-limited even if you wipe your neovim cache.

As to why it still works in the browser, I suspect they rate-limit the API separately to their shiny UI (it would probably be hard to give the web UI such heavy usage as the API anyway, and in the web UI they control and probably optimise what gets sent to the foundational models).

Are you using suggestions? That got me rate-limited with free Copilot.