Inline Assistant

​

Activating Inline Assist

1. Select code (or place cursor for insertion)
2. Press Cmd+Enter (macOS) or Ctrl+Enter (Linux/Windows)
3. Type your natural language prompt
4. Press Enter to generate

​

Use Cases

​

Code Generation

1. Place cursor where you want code inserted
2. Cmd+Enter → “Create a function that validates email addresses”
3. Review the generated code
4. Accept with Tab or reject with Esc

​

Code Transformation

1. Select the code to transform
2. Cmd+Enter → “Convert this to async/await”
3. Review the diff
4. Accept or modify

"Add error handling"
"Convert to TypeScript"
"Extract this into a separate function"
"Optimize for performance"
"Add JSDoc comments"

​

Multi-selection Editing

1. Create multiple selections (Cmd+D to select next occurrence)
2. Cmd+Enter → “Add null checks”
3. AI processes each selection independently
4. Review all changes together

​

Refactoring

"Convert this class to use dependency injection"
"Split this large function into smaller helpers"
"Replace this with the Builder pattern"

​

Documentation

"Add comprehensive JSDoc for this API"
"Write a doc comment explaining the algorithm"
"Generate usage examples for this function"

​

Inline Assist UI

​

Prompt Editor

• Input field - Type your natural language instruction
• Model selector - Choose which AI model to use
• Settings icon - Access inline assistant settings
• Context indicators - Shows what context is being sent

​

Generation Preview

• Streaming text - See code appear in real-time
• Diff highlighting -
  • Green: Added lines
  • Red: Removed lines
  • Modified lines show both colors
• Thinking indicator - Shows when model is reasoning (for thinking-enabled models)

​

Action Buttons

• Stop (Esc) - Cancel generation

• Accept (Tab) - Apply changes
• Reject (Esc) - Discard changes
• Retry (Cmd+R) - Generate a new alternative
• Edit prompt (Cmd+E) - Modify your instruction

​

Alternative Suggestions

​

Configure Alternative Models

{
  "agent": {
    "inline_alternatives": [
      {
        "provider": "anthropic",
        "model": "claude-3-5-sonnet-20241022"
      },
      {
        "provider": "openai",
        "model": "gpt-4o"
      }
    ]
  }
}

​

Cycling Alternatives

• Press Cmd+R (macOS) or Ctrl+R (Linux/Windows) to generate a new alternative
• Use arrow buttons to cycle through generated alternatives
• Each alternative is generated by a different model or with different parameters

​

Advanced Features

​

Context Integration

• Selected code and surrounding context
• File type and language information
• Imports and dependencies at the top of the file
• LSP information (type definitions, symbol information)
• Diagnostic errors in the selected region

​

Custom Instructions

Use the lodash library for this transformation

Write this following our coding standards:
- Use const/let, not var
- Add TypeScript types
- Include error handling

​

Code Actions Integration

1. Place cursor on a diagnostic (error/warning)
2. Cmd+. (macOS) or Ctrl+. (Linux/Windows)
3. Select “Ask AI to fix this”

​

Configuration

​

Model Selection

{
  "agent": {
    "inline_assistant_model": {
      "provider": "openai",
      "model": "gpt-4o"
    }
  }
}

​

Streaming Tools

{
  "agent": {
    "inline_assistant_use_streaming_tools": true
  }
}

• The AI can make multiple edits in one generation
• Faster iteration on complex transformations
• Better handling of multi-file changes

​

Temperature Override

{
  "agent": {
    "model_parameters": [
      {
        "provider": "anthropic",
        "model": "claude-3-5-sonnet-20241022",
        "temperature": 0.7
      }
    ]
  }
}

​

Keyboard Shortcuts

​

Best Practices

​

Be Specific

​

Provide Context

​

Use Natural Language

• “Make this code safer” ✓
• “Add error handling for network failures” ✓
• “Use the same pattern as the login function” ✓

​

Iterate

1. Generate initial code
2. Accept it
3. Select what needs improvement
4. Cmd+Enter → “Now add logging”

​

Review Before Accepting

• Check for logic errors
• Verify it matches your coding style
• Ensure dependencies exist
• Test the changes

​

Combine with Editor Features

• Use multi-cursor for repeated patterns
• Select code blocks (functions, classes) for targeted changes
• Leverage language server errors to guide fixes

​

Examples

​

Example 1: Add Error Handling

async function fetchUser(id: string) {
  const response = await fetch(`/api/users/${id}`);
  return response.json();
}

async function fetchUser(id: string) {
  try {
    const response = await fetch(`/api/users/${id}`);
    
    if (!response.ok) {
      throw new Error(`Failed to fetch user: ${response.statusText}`);
    }
    
    return await response.json();
  } catch (error) {
    console.error('Error fetching user:', error);
    throw error;
  }
}

​

Example 2: Convert to TypeScript

function calculateTotal(items) {
  return items.reduce((sum, item) => sum + item.price * item.quantity, 0);
}

interface CartItem {
  price: number;
  quantity: number;
}

function calculateTotal(items: CartItem[]): number {
  return items.reduce((sum, item) => sum + item.price * item.quantity, 0);
}

​

Example 3: Generate Tests

fn parse_email(input: &str) -> Result<String, ParseError> {
    // implementation
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_parse_valid_email() {
        assert_eq!(
            parse_email("[email protected]").unwrap(),
            "[email protected]"
        );
    }

    #[test]
    fn test_parse_invalid_email() {
        assert!(parse_email("invalid").is_err());
    }

    #[test]
    fn test_parse_empty_string() {
        assert!(parse_email("").is_err());
    }
}

​

Troubleshooting

​

Generation is slow

• Try a faster model (e.g., GPT-4o-mini instead of GPT-4o)
• Reduce the selected context size
• Check your internet connection
• Consider using a local model via Ollama

​

Results are not relevant

• Be more specific in your prompt
• Include more surrounding code in your selection
• Mention specific libraries or patterns to use
• Try a different model

​

Changes are not being applied

• Make sure you press Tab to accept
• Check that the file is not read-only
• Verify the file hasn’t been modified externally

​

Context is wrong

• The assistant uses the code around your selection
• Expand your selection to include more context
• Mention specific files or symbols in your prompt