Skip to main content
Avante.nvim is a Neovim plugin designed to emulate the behavior of AI-powered coding assistants like Cursor, bringing sophisticated AI-driven code suggestions and autonomous code editing directly into your Neovim environment.

​
Design Philosophy

Avante.nvim is built on three core principles:
  1. Native Neovim Integration: Deep integration with Neovim’s architecture, leveraging its buffer system, UI components, and plugin ecosystem
  2. Flexibility: Support for multiple AI providers and interaction modes to suit different workflows
  3. Transparency: Clear visibility into AI operations with tool execution logs, permission controls, and prompt logging

​
Core Components

Avante.nvim’s architecture consists of several key components working together: The sidebar (lua/avante/sidebar.lua) is the primary interface for interacting with Avante. It manages:
  • Result display: Shows AI responses and code suggestions
  • Input handling: Captures user prompts and commands
  • File selection: Manages context files for AI conversations
  • History management: Maintains conversation history across sessions

​
LLM Engine

The LLM engine (lua/avante/llm.lua) handles all AI provider interactions:
  • Manages streaming responses from AI providers
  • Coordinates tool execution and autonomous workflows
  • Handles conversation history and memory management
  • Supports both traditional chat and agentic modes
-- Example: Basic LLM configuration
{
  provider = "claude",
  mode = "agentic",  -- or "legacy"
  providers = {
    claude = {
      model = "claude-sonnet-4-5-20250929",
      timeout = 30000,
      extra_request_body = {
        temperature = 0.75,
        max_tokens = 64000,
      },
    },
  },
}

​
Provider System

The provider system (lua/avante/providers/) abstracts different AI services:
  • Supported providers: Claude, OpenAI, Gemini, Copilot, Azure, and many more
  • Custom providers: Extensible architecture for adding new AI services
  • ACP providers: Special providers using the Agent Client Protocol for enhanced capabilities
See Providers for detailed information.

​
Tool System

The tool system (lua/avante/llm_tools/) enables autonomous code editing:
  • File operations (read, write, edit, create)
  • Code search and navigation (grep, glob, diagnostics)
  • Shell command execution (bash)
  • Task management (write_todos, read_todos)
  • Version control (undo_edit)
See Agentic Workflow for more details on tools.

​
Interaction Modes

Avante supports two primary interaction modes:
The modern, autonomous approach where the AI can:
  • Execute tools to read and modify files
  • Search through your codebase
  • Run shell commands
  • Autonomously implement changes
Best for: Complex refactoring, feature implementation, multi-file changes
-- Switch between modes in config
{
  mode = "agentic",  -- Default: autonomous tools
  -- or
  mode = "legacy",   -- Traditional chat-based
}
Learn more about modes in Modes.

​
Key Features

​
Project Context

Avante can understand your project through:
  • File selection: Explicitly add files to conversation context
  • Repo mapping: Automatic codebase structure analysis
  • Diagnostics: Integration with LSP diagnostics
  • RAG service: Optional semantic search over your codebase

​
Permission System

Control what AI can do in your project: 
{
  behaviour = {
    -- Auto-approve all tools (fastest)
    auto_approve_tool_permissions = true,
    
    -- Prompt for every tool (most controlled)
    -- auto_approve_tool_permissions = false,
    
    -- Auto-approve specific tools only
    -- auto_approve_tool_permissions = {"bash", "str_replace"},
  },
}

​
History & Memory

Avante maintains conversation context across sessions:
  • Chat history stored per-project
  • Automatic memory summarization for long conversations
  • History selector to resume previous conversations
  • Token counting to manage context limits

​
Configuration Structure

Avante’s configuration follows a hierarchical structure: 
require('avante').setup({
  -- Core settings
  provider = "claude",
  mode = "agentic",
  
  -- Provider configurations
  providers = {
    -- Each provider has its own config
  },
  
  -- Behavior settings
  behaviour = {
    auto_suggestions = false,
    auto_apply_diff_after_generation = false,
    -- ... more options
  },
  
  -- UI settings
  windows = {
    position = "right",
    width = 30,
    -- ... more options
  },
  
  -- Key mappings
  mappings = {
    -- Customizable keybindings
  },
})
For complete configuration options, see the Configuration Reference.

​
Next Steps

Modes

Learn about agentic vs legacy modes

Providers

Understand AI provider architecture

Agentic Workflow

Deep dive into autonomous tools

Zen Mode

CLI-like experience in Neovim

Build docs developers (and LLMs) love

Get started for freeTalk to us