GitNexus/DEBUG_FEATURES.md

🔍 Debug Mode Features

The GitNexus chat interface now includes a comprehensive debug mode that shows the internal workings of the Graph RAG agent.

📝 Markdown Formatting

NEW: The chat interface now supports full markdown formatting for better readability!

Supported Markdown Features:

• Headers (# ## ###) for organizing information
• Bold and italic text for emphasis
• Inline code for function names and file paths
• Code blocks with syntax highlighting for multiple languages
• Bullet points and numbered lists
• Tables for structured data
• Blockquotes for important notes
• Links (automatically open in new tabs)

Enhanced Debug Display:

• Reasoning observations are now rendered with markdown
• Query explanations support formatted text
• Tool outputs preserve formatting and structure
• Code snippets get proper syntax highlighting

How to Use Debug Mode

1. Toggle Debug Mode: Click the 🔍 Debug button in the chat interface header
2. Ask Questions: When debug mode is enabled, all assistant responses will include detailed debug information
3. Explore Tabs: The debug panel includes multiple tabs showing different aspects of the processing

Debug Panel Tabs

🧠 Reasoning Steps

Shows the complete ReAct (Reasoning + Acting) process:

• Step-by-step thinking: See how the agent reasons about your question
• Actions taken: View which tools the agent decides to use
• Tool inputs: See the exact parameters passed to each tool
• Observations: View the results returned by each tool
• Success/failure status: Monitor tool execution success

🔍 Cypher Queries

Displays generated graph queries:

• Generated Cypher: See the exact database queries created
• Query explanations: Understand why each query was generated
• Confidence scores: View how confident the system is in each query
• Syntax highlighting: Cypher queries displayed with proper formatting

⚙️ Configuration

Shows system configuration:

• LLM Settings: Provider, model, temperature, max tokens
• RAG Options: Reasoning steps, strict mode, temperature
• Performance Metrics: Execution time, confidence scores

📊 Context Info

Displays knowledge graph statistics:

• Graph Nodes: Total number of code entities in the graph
• Files Indexed: Number of source files processed
• Sources Used: Files referenced in the current response
• Referenced Sources: List of specific files used for the answer

What You Can Learn

Understanding Agent Behavior

• See how the agent breaks down complex questions
• Understand the reasoning process step-by-step
• Monitor which tools are used and why

Query Optimization

• View generated Cypher queries to understand graph traversal
• See query confidence scores to assess reliability
• Learn about query patterns for different question types

Performance Analysis

• Monitor execution times for different operations
• Understand the relationship between question complexity and processing time
• Identify bottlenecks in the reasoning process

Context Awareness

• See how much of your codebase is being used
• Understand which files are most relevant to your questions
• Monitor the scope of graph traversal

Debug Mode Benefits

1. Transparency: Complete visibility into AI decision-making
2. Learning: Understand how Graph RAG works internally
3. Debugging: Identify issues with queries or reasoning
4. Optimization: Fine-tune your questions for better results
5. Trust: Build confidence through explainable AI

Example Debug Output

When you ask "How many functions are in this project?", debug mode shows:

Reasoning Steps:

1. Thought: "I need to count all functions in the project using a graph query"
2. Action: query_graph
3. Input: "Count all functions in the project"
4. Observation: Generated Cypher query and results

Generated Query:

MATCH (f:Function) RETURN COUNT(f)

Configuration:

• Model: gpt-4o-mini
• Temperature: 0.1
• Execution Time: 1,234ms

This level of detail helps you understand exactly how your question was processed and answered.