avante.nvim

The dual_boost feature is an experimental mode that generates two separate responses from two different providers (first_provider and second_provider) and then uses a third prompt to combine them into a single response.

Configuration keys:

• enabled: Boolean to enable the feature.
• first_provider: The first LLM to call.
• second_provider: The second LLM to call.
• prompt: The template used to merge outputs. It supports {{provider1_output}} and {{provider2_output}} placeholders.
• timeout: Timeout in milliseconds.

dual_boost = {
  enabled = false,
  first_provider = "openai",
  second_provider = "claude",
  prompt = "Based on the following two reference outputs, generate a response that combines elements of both but reflects your own judgment and unique perspective. Do not provide any explanation, just give the response directly. Reference output 1: [{{provider1_output}}], Reference output 2: [{{provider2_output}}]",
  timeout = 60000,
}

If you use blink.cmp, you can integrate Avante's completion sources (Mentions, Slash Commands, Shortcuts, and Files) using blink.compat or the Kaiser-Yang/blink-cmp-avante plugin.

Available Completion Sources:

• Mentions (@): @codebase, @diagnostics, @file, @quickfix, @buffers.
• Slash Commands (/): /help, /init, /clear, /new, /compact, /lines <start>-<end> <question>, /commit.
• Shortcuts (#): Predefined prompt templates.
• Files: File selection for context.

-- Example blink.cmp configuration with Avante sources
      default = {
        ...,
        "avante_commands",
        "avante_mentions",
        "avante_shortcuts",
        "avante_files",
      }

      providers = {
        avante_commands = {
          name = "avante_commands",
          module = "blink.compat.source",
          score_offset = 90,
          opts = {},
        },
        avante_files = {
          name = "avante_files",
          module = "blink.compat.source",
          score_offset = 100,
          opts = {},
        },
        avante_mentions = {
          name = "avante_mentions",
          module = "blink.compat.source",
          score_offset = 1000,
          opts = {},
        },
        avante_shortcuts = {
          name = "avante_shortcuts",
          module = "blink.compat.source",
          score_offset = 1000,
          opts = {},
        }
        ...
      }

The recommended way to install avante.nvim is via lazy.nvim. Note that you must include the build step. If you want to build from source, you must set BUILD_FROM_SOURCE=true in your make command (though the provided snippet uses a conditional for Windows/Unix).

Important Requirements:

• Neovim 0.11.0 or later is required.
• If building from source, cargo is required.
• For Windows users, powershell is used for the build process.

Recommended Neovim Option: To ensure the sidebar can be fully collapsed, set vim.opt.laststatus = 3 in your Neovim configuration.

{
  "yetone/avante.nvim",
  build = vim.fn.has("win32") ~= 0
      and "powershell -ExecutionPolicy Bypass -File Build.ps1 -BuildFromSource false"
      or "make",
  event = "VeryLazy",
  version = false, -- Never set this value to "*"!
  opts = {
    provider = "claude",
    -- add other opts here
  },
  dependencies = {
    "nvim-lua/plenary.nvim",
    "MunifTanjim/nui.nvim",
    -- optional dependencies
    "nvim-telescope/telescope.nvim",
    "hrsh7th/nvim-cmp",
    "folke/snacks.nvim",
    -- ...
  },
}

Avante supports the Agent Client Protocol (ACP), allowing integration with AI agents that use standardized JSON-RPC communication. ACP providers are configured in the acp_providers section. Supported agents include gemini-cli, claude-code, goose, codex, and kimi-cli.

Prerequisites:

• Gemini CLI: Install gemini and set GEMINI_API_KEY.
• Claude Code: Install acp-claude-code via npm and set ANTHROPIC_API_KEY.

acp_providers = {
  ["gemini-cli"] = {
    command = "gemini",
    args = { "--experimental-acp" },
    env = {
      NODE_NO_WARNINGS = "1",
      GEMINI_API_KEY = os.getenv("GEMINI_API_KEY"),
    },
  },
  ["claude-code"] = {
    command = "claude-agent-acp",
    args = { },
    env = {
      NODE_NO_WARNINGS = "1",
      ANTHROPIC_API_KEY = os.getenv("ANTHROPIC_API_KEY"),
    },
  },
  ["goose"] = {
    command = "goose",
    args = { "acp" },
  },
  ["codex"] = {
    command = "codex-acp",
    args = {},
    env = {
      NODE_NO_WARNINGS = "1",
      OPENAI_API_KEY = os.getenv("OPENAI_API_KEY"),
    },
  },
}