AI Providers

Understanding AI providers in Avante.nvim and how they power your coding assistant

What are Providers?

Providers are the AI services that power Avante.nvim’s intelligent code assistance. Each provider offers different models, capabilities, and pricing structures, giving you the flexibility to choose the best option for your needs.

Provider Architecture

Avante.nvim uses a flexible provider system that allows you to:

• Switch between providers on the fly
• Use different providers for different tasks
• Create custom providers to integrate with any AI service
• Configure provider-specific settings and models

How Providers Work

-- Provider configuration structure
providers = {
  provider_name = {
    endpoint = "https://api.example.com",
    model = "model-name",
    api_key_name = "ENV_VAR_NAME",
    parse_curl_args = function(opts, code_opts)
      -- Custom request formatting
    end,
    parse_response_data = function(data_stream, event_state, opts)
      -- Custom response parsing
    end,
  },
}

Available Providers

Avante.nvim supports a wide range of AI providers out of the box:

Claude

Anthropic’s Claude models with extended context and reasoning

OpenAI

GPT-4o and reasoning models (o1, o3-mini)

Gemini

Google’s Gemini with 1M+ token context window

GitHub Copilot

Use your Copilot subscription with Avante

Ollama

Run local models privately on your machine

Custom Providers

Create your own provider integrations

Provider Selection

Default Provider

Set your default provider in the configuration:

require('avante').setup({
  provider = "claude",  -- Default provider
})

Specialized Providers

You can use different providers for different tasks:

require('avante').setup({
  provider = "claude",                    -- Main provider
  auto_suggestions_provider = "copilot",  -- Fast suggestions
  memory_summary_provider = "openai",     -- Memory summaries
})

Since auto-suggestions are a high-frequency operation, using expensive providers like Copilot can be costly. Consider using a local model with Ollama for suggestions.

Runtime Switching

Switch providers while Neovim is running:

:AvanteSwitchProvider claude
:AvanteSwitchProvider openai
:AvanteSwitchProvider ollama

Or use the interactive picker:

:AvanteSwitchProvider

Provider Capabilities

Streaming Responses

All built-in providers support streaming responses, allowing you to see AI output as it’s generated:

providers = {
  custom = {
    parse_response_data = function(data_stream, event_state, opts)
      -- Handle streaming data
      if data_stream:match('"delta":') then
        -- Extract and return delta content
      end
    end,
  },
}

Tool Calling

Providers that support tool calling enable agentic workflows:

providers = {
  claude = {
    -- Tools are automatically supported
    -- Disable specific tools if needed
    __inherited_tools = { "bash", "str_replace" },
  },
}

Context Windows

Different providers have different context window sizes:

Provider	Context Window	Notes
Claude Sonnet 4.5	200K tokens	Prompt caching available
GPT-4o	128K tokens	Structured outputs
Gemini 2.0 Flash	1M+ tokens	Massive context
Ollama (varies)	8K-128K tokens	Model dependent

API Keys and Authentication

Each provider requires authentication. Avante supports both scoped and global API keys:

Scoped API Keys (Recommended)

# Avante-specific keys (won't affect other tools)
export AVANTE_ANTHROPIC_API_KEY=your-claude-key
export AVANTE_OPENAI_API_KEY=your-openai-key
export AVANTE_GEMINI_API_KEY=your-gemini-key

Global API Keys

# System-wide keys
export ANTHROPIC_API_KEY=your-claude-key
export OPENAI_API_KEY=your-openai-key
export GEMINI_API_KEY=your-gemini-key

Avante will check for scoped keys first, then fall back to global keys.

Provider Inheritance

Custom providers can inherit from existing providers:

providers = {
  my_custom_claude = {
    __inherited_from = "claude",
    model = "claude-opus-4-20250514",
    endpoint = "https://my-proxy.com/v1",
  },
}

This inherits all the parsing logic while allowing customization.

Next Steps

Configure Providers

Detailed provider configuration options

Claude Setup

Get started with Claude

Ollama Local Models

Run AI models locally

Custom Providers

Create your own provider