> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/yetone/avante.nvim/llms.txt
> Use this file to discover all available pages before exploring further.

# Auto-Suggestions

> Inline AI code completions and suggestions

## Overview

Avante.nvim provides auto-suggestions (also known as inline completions) that give you AI-powered code suggestions as you type, similar to GitHub Copilot. This feature is currently in experimental stage.

<Warning>
  Auto-suggestions are a high-frequency operation and can be expensive. Be mindful of your API usage, especially with providers like Copilot.
</Warning>

## Enabling Auto-Suggestions

Auto-suggestions are disabled by default. Enable them in your configuration:

```lua theme={null}
require('avante').setup({
  behaviour = {
    auto_suggestions = true, -- Enable auto-suggestions
  },
  auto_suggestions_provider = "claude", -- Provider to use
})
```

### Provider Selection

<Warning>
  Designating `copilot` as the auto-suggestions provider can be dangerous due to high-frequency requests. See [issue #1048](https://github.com/yetone/avante.nvim/issues/1048) for details.
</Warning>

Recommended providers for auto-suggestions:

* `"claude"` - Claude models (recommended)
* `"openai"` - OpenAI models
* `"gemini"` - Google Gemini

Avoid using `"copilot"` unless you understand the implications.

## Configuration

### Debounce and Throttle Settings

To control the frequency of suggestion requests, configure debounce and throttle:

```lua theme={null}
require('avante').setup({
  suggestion = {
    debounce = 600,  -- Wait 600ms after typing stops
    throttle = 600,  -- Maximum one request per 600ms
  },
})
```

<ParamField path="debounce" type="number" default="600">
  Time in milliseconds to wait after typing stops before requesting a suggestion
</ParamField>

<ParamField path="throttle" type="number" default="600">
  Minimum time in milliseconds between suggestion requests
</ParamField>

<Tip>
  If you're using the Copilot provider or experiencing too many API calls, increase these values to reduce request frequency:

  ```lua theme={null}
  suggestion = {
    debounce = 1000,
    throttle = 1000,
  }
  ```
</Tip>

## Keybindings

Default keybindings for working with suggestions:

| Key Binding | Mode   | Description                |
| ----------- | ------ | -------------------------- |
| `<M-l>`     | Insert | Accept current suggestion  |
| `<M-]>`     | Insert | Show next suggestion       |
| `<M-[>`     | Insert | Show previous suggestion   |
| `<C-]>`     | Insert | Dismiss current suggestion |

<Info>
  `<M-l>` means Alt+l on Linux/Windows or Option+l on macOS.
</Info>

### Customizing Keybindings

You can customize the suggestion keybindings in your configuration:

```lua theme={null}
require('avante').setup({
  mappings = {
    suggestion = {
      accept = "<M-l>",      -- Accept suggestion
      next = "<M-]>",        -- Next suggestion
      prev = "<M-[>",        -- Previous suggestion
      dismiss = "<C-]>",     -- Dismiss suggestion
    },
  },
})
```

### Custom Keybinding Example

```lua theme={null}
-- Use Tab to accept, Ctrl+n/p for navigation
require('avante').setup({
  mappings = {
    suggestion = {
      accept = "<Tab>",
      next = "<C-n>",
      prev = "<C-p>",
      dismiss = "<Esc>",
    },
  },
})
```

## Toggle Suggestions

You can toggle the suggestion display on and off:

```vim theme={null}
:AvanteToggleSuggestion
```

Or use the default keybinding:

```vim theme={null}
<Leader>as
```

## Provider Configuration

Configure your auto-suggestions provider with appropriate settings:

```lua theme={null}
require('avante').setup({
  auto_suggestions_provider = "claude",
  providers = {
    claude = {
      endpoint = "https://api.anthropic.com",
      model = "claude-3-5-sonnet-20241022",
      timeout = 30000,
      extra_request_body = {
        temperature = 0.5, -- Lower temperature for more focused suggestions
        max_tokens = 2048,
      },
    },
  },
})
```

## How Auto-Suggestions Work

<Steps>
  <Step title="Typing Detection">
    As you type, Avante monitors your input and waits for the debounce period.
  </Step>

  <Step title="Context Gathering">
    When the debounce period expires, Avante gathers context from:

    * Current file content
    * Cursor position
    * Recent changes
    * Project context (if RAG is enabled)
  </Step>

  <Step title="Request Generation">
    A suggestion request is sent to the configured provider (respecting throttle limits).
  </Step>

  <Step title="Inline Display">
    The suggestion is displayed inline as ghost text, which you can accept, cycle through, or dismiss.
  </Step>
</Steps>

## Best Practices

<CardGroup cols={2}>
  <Card title="Choose the Right Provider" icon="plug">
    Use providers with good latency and reasonable pricing for auto-suggestions. Claude is generally recommended.
  </Card>

  <Card title="Adjust Timing" icon="clock">
    Fine-tune `debounce` and `throttle` to balance responsiveness with API costs.
  </Card>

  <Card title="Monitor Usage" icon="chart-line">
    Keep an eye on your API usage, especially during heavy coding sessions.
  </Card>

  <Card title="Use with RAG" icon="database">
    Enable the RAG service for better context-aware suggestions.
  </Card>
</CardGroup>

## Troubleshooting

<AccordionGroup>
  <Accordion title="Suggestions are too slow">
    1. Reduce `debounce` value (but this increases API calls)
    2. Use a faster provider (e.g., Claude Haiku instead of Sonnet)
    3. Check your internet connection
    4. Ensure your provider endpoint is responsive
  </Accordion>

  <Accordion title="Too many API calls">
    1. Increase `debounce` value (e.g., to 1000ms)
    2. Increase `throttle` value
    3. Consider disabling auto-suggestions for certain file types
  </Accordion>

  <Accordion title="Suggestions aren't appearing">
    1. Verify `behaviour.auto_suggestions = true`
    2. Check that your provider is configured correctly
    3. Ensure API keys are set
    4. Look for errors in `:messages`
  </Accordion>

  <Accordion title="Suggestions are low quality">
    1. Enable the RAG service for better context
    2. Adjust provider temperature (lower for more focused)
    3. Ensure you're using a capable model
    4. Add project instructions via `avante.md`
  </Accordion>
</AccordionGroup>

## Disabling Auto-Suggestions

If you want to disable auto-suggestions:

```lua theme={null}
require('avante').setup({
  behaviour = {
    auto_suggestions = false,
  },
})
```

Or toggle them temporarily with `<Leader>as` or `:AvanteToggleSuggestion`.

## Related Documentation

<CardGroup cols={2}>
  <Card title="Providers" icon="plug" href="/configuration/providers">
    Configure AI providers
  </Card>

  <Card title="Keybindings" icon="keyboard" href="/configuration/keybindings">
    Customize all keybindings
  </Card>

  <Card title="RAG Service" icon="database" href="/features/rag-service">
    Enable RAG for better suggestions
  </Card>

  <Card title="Project Instructions" icon="file" href="/configuration/project-instructions">
    Guide AI with project-specific context
  </Card>
</CardGroup>