An AI-powered Language Server that provides intelligent semantic code analysis beyond what traditional LSPs can detect. Built with modern Python tools and designed to integrate seamlessly with any LSP-compatible editor.

AI LSP focuses exclusively on semantic issues that require deep understanding:

• Logic errors in algorithms and business logic
• Subtle race conditions and concurrency issues
• Architectural problems and design pattern violations
• Security vulnerabilities requiring context understanding
• Performance anti-patterns needing semantic analysis
• Complex data flow issues
• Domain-specific best practice violations
• Accessibility issues requiring UX understanding

What it doesn't do: Traditional LSP tasks like syntax errors, type errors, undefined variables, or basic formatting - your regular LSP handles those perfectly.

• 🧠 AI-Powered Analysis: Uses advanced language models for semantic code understanding
• ⚡ Debounced Analysis: Smart delays prevent AI spam during rapid typing
• 🔧 Code Actions: Provides automated fixes for detected issues
• 🎯 Multi-Language: Supports Python, JavaScript, TypeScript, Rust, Go, Java, C/C++, and Lua
• ⚖️ Async Architecture: Non-blocking analysis with race condition protection

uv add ai-lsp
# or
pip install ai-lsp

git clone https://github.com/benomahony/ai-lsp
cd ai-lsp
uv sync

Set your AI provider API key:

# For Google AI (default)
export GOOGLE_API_KEY="your-api-key"

# Or for OpenAI
export OPENAI_API_KEY="your-openai-key" 

Configure via environment variables or .env file:

AI_LSP_MODEL=google-gla:gemini-2.5-pro  # Default model
DEBOUNCE_MS=1000                         # Analysis delay (milliseconds)

Supported models:

• google-gla:gemini-2.5-pro (default)
• gpt-4o
• Any model supported by pydantic-ai

Create ~/.config/nvim/lua/plugins/ai-lsp.lua:

return {
  {
    "neovim/nvim-lspconfig",
    opts = {
      servers = {
        ["ai-lsp"] = {
          mason = false,
          cmd = { "ai-lsp" },
          filetypes = { "python", "javascript", "typescript", "rust", "go", "lua", "java", "cpp", "c" },
        },
      },
    },
  },
}

Add to your settings.json:

{
  "ai-lsp.command": ["uvx", "ai-lsp"],
  "ai-lsp.filetypes": ["python", "javascript", "typescript", "rust", "go", "lua", "java", "cpp", "c"]
}

Or if you have it installed in your active environment:

{
  "ai-lsp.command": ["ai-lsp"],
  "ai-lsp.filetypes": ["python", "javascript", "typescript", "rust", "go", "lua", "java", "cpp", "c"]
}

Any LSP client can use AI LSP with these settings:

{
  "command": ["uvx", "ai-lsp"],
  "filetypes": ["python", "javascript", "typescript", "rust", "go", "lua", "java", "cpp", "c"],
  "rootPatterns": [".git", "pyproject.toml", "package.json", "Cargo.toml", "go.mod"]
}

Given this Python code:

def authenticate(username, password):
    if password == 'admin123':  # AI LSP will flag this
        print(f'User {username} authenticated')  # And suggest logging
        return True
    else:
        print('Authentication failed')
        return False

AI LSP might detect:

1. Security Issue: Hardcoded password 'admin123'

   • Severity: Error
   • Suggestion: Use environment variables or secure credential storage

2. Best Practice: Using print statements

   • Severity: Warning
   • Code Action: Replace with logging.info()

LSP not starting:

• Verify ai-lsp command is in PATH: which ai-lsp
• Check log file for errors (default: ai-lsp.log)
• Ensure API keys are set correctly

No diagnostics appearing:

• Check file type is supported
• Verify AI model has API access
• Look for errors in log file

Performance issues:

• Increase DEBOUNCE_MS to reduce API calls
• Use a faster model like gpt-4o-mini
• Check network connectivity to AI provider

Enable verbose logging:

ai-lsp --log-level DEBUG --log-file debug.log

MIT License - see LICENSE for details.