LANGCHAIN_ROUTER.md

LangChain Command Router Implementation

Overview

The desktop assistant now uses LangChain and LLM-based intelligent intent classification instead of regex pattern matching for command routing.

What Changed

Before (Regex-based)

m_create_note = re.match(r'^\s*create\s+note\b(.*)$', user_input_lower)
m_send_email = re.match(r'^\s*send\s+email\b(.*)$', user_input_lower)

if m_create_note:
    # handle create note
elif m_send_email:
    # handle send email

After (LangChain-based)

classification = command_router.classify_intent(user_input)
command = classification['command']

if command == 'create_note':
    # handle create note
elif command == 'send_email':
    # handle send email
elif command == 'conversation':
    # handle general conversation

Benefits

1. Flexible Input: Users don't need to say exact command phrases

   • "make a note about the meeting" → create_note
   • "email john about the project" → send_email
   • "do i have any meetings" → check_gmail

2. Intelligent Classification: LLM understands intent contextually

   • Distinguishes between commands and conversations
   • Extracts parameters automatically
   • Provides confidence scores

3. Natural Conversations: The system knows when users want to chat vs execute commands

   • "how are you" → conversation
   • "tell me a joke" → conversation
   • "create meeting tomorrow" → create_event

Architecture

CommandRouter (core/command_router.py)

• Uses LLM to classify user intent
• Returns structured classification with:
  • intent: "command" or "conversation"
  • command: The specific command name
  • parameters: Extracted information
  • confidence: Classification confidence (0-1)
• Fallback to regex if LLM fails

Main Loop (main_with_langchain.py)

• Calls command_router.classify_intent(user_input)
• Routes to appropriate handler based on command
• Handles both specific commands and general conversation

Supported Commands

Command	Description	Examples
create_note	Create a new note	"create note meeting notes", "make a note"
delete_note	Delete a note	"delete note old tasks", "remove the note"
list_notes	List all notes	"list notes", "show my notes"
send_email	Send an email	"send email to john", "email sarah"
check_gmail	Check Gmail	"check gmail for meetings", "any emails"
create_event	Create calendar event	"create event tomorrow at 3pm"
list_events	List events	"list events for next week"
search_event	Search events	"search event about project"
conversation	General chat	"how are you", "tell me a joke"
exit	Exit application	"exit", "quit", "goodbye"

Testing

Run the test script to verify the command router:

python test_command_router.py

This will test various inputs and show how they're classified.

Dependencies

Added to requirements.txt:

• langchain - Core LangChain library
• langchain-community - Community integrations
• sentence-transformers - Already included for Gmail similarity search

Install with:

pip install -r requirements.txt

Usage

Simply run the main file:

python "main_with_langchain .py"

The system will automatically classify your intent and route to the appropriate handler. No changes to voice commands needed - it's more flexible than before!

Customization

To add new commands:

1. Add command to CommandRouter.commands dict in core/command_router.py
2. Add handler in main loop (main_with_langchain.py)
3. Update this documentation

Example:

# In command_router.py
'open_app': {
    'description': 'Open an application',
    'examples': ['open chrome', 'launch spotify']
}

# In main_with_langchain.py
elif command == 'open_app':
    # Handle app opening
    pass