Show HN: Triage.flow – Chat with Any GitHub Repo Using Faiss and LlamaIndex

Agentic AI-Powered Interactive GitHub Repository Analysis with Advanced Chat Interface

A modern, web-based agentic AI assistant that helps you understand, analyze, and explore entire GitHub repositories through an interactive chat interface. Powered by ReAct (Reasoning + Acting) agents that autonomously use tools, reason through complex problems, and provide deep, code-aware insights with beautiful presentation, smart context retrieval, and cost-optimized AI interactions.

• 🤖 Agentic AI System - ReAct agents that think, plan, and use tools autonomously to solve complex coding questions
• 🧠 Real-time Reasoning - Watch the AI think through problems with live Thought/Action/Observation streaming
• 🛠️ Autonomous Tool Usage - AI independently explores files, searches code, and analyzes patterns across your repository
• 🎯 Interactive Repository Chat - ChatGPT-style conversation with full repository context and agent reasoning
• 🌲 Smart Codebase Explorer - High-performance file picker with search and tree navigation
• 📋 Enhanced Markdown - Context-aware emojis, file linking, and professional formatting
• 📎 Smart File & Folder Mentions - Use @filename.py or @folder/path with autocomplete to reference specific files/folders
• 👀 Integrated File Viewer - Side-by-side file viewing pane with syntax highlighting and quick actions
• ⚡ Real-time Streaming - Live AI responses with typing indicators and cost optimization
• 🎨 Modern UI/UX - Professional, responsive interface with session management
• 🔍 Advanced RAG System - Smart context sizing with 10-25 sources from indexed repositories
• 💰 Cost-Optimized AI - OpenRouter prompt caching for 25-90% cost savings

Our system uses ReAct (Reasoning + Acting) agents that don't just answer questions - they autonomously reason through complex problems and take actions to solve them:

• Multi-step problem solving - Breaks down complex questions into logical steps
• Dynamic tool selection - Chooses the right tools for each task automatically
• Context-aware decisions - Adapts approach based on repository structure and user intent
• Self-correction - Re-evaluates and adjusts when initial approaches don't work

• File exploration - Autonomously navigates and examines repository structure
• Smart code search - Semantically searches across files for relevant patterns
• Pattern analysis - Identifies architectural patterns and relationships
• Code generation - Creates context-aware examples based on existing codebase patterns

• Live reasoning display - See exactly how the AI thinks through problems
• Tool execution tracking - Watch as tools are selected and executed
• Step-by-step breakdown - Understand the agent's decision-making process
• Interactive feedback - Agent adapts based on your responses and clarifications

• Frontend: React 18, TypeScript, Tailwind CSS, shadcn/ui
• Backend: Python FastAPI, FAISS Vector Store, Advanced RAG
• AI Models: OpenAI, OpenRouter (Claude, Mistral, Llama) with prompt caching
• Code Analysis: Tree-sitter parsers for 20+ languages with smart fallbacks
• Real-time: Server-Sent Events for streaming responses
• Performance: Multi-level caching, smart context sizing, async processing

• LlamaIndex-powered agents with autonomous tool selection
• Multi-step reasoning with self-correction capabilities
• Real-time streaming of thought processes and tool executions
• Context-aware memory that remembers previous interactions and reasoning chains

• explore_directory - Autonomous file system navigation with metadata
• search_codebase - Semantic code search across entire repositories
• read_file - Smart file reading with size optimization
• analyze_file_structure - Deep structural analysis and pattern recognition
• find_related_files - Relationship discovery through imports and naming
• semantic_content_search - AI-powered content understanding
• generate_code_example - Context-aware code generation from repository patterns

The UI provides real-time insight into the agent's decision-making:

• 🧠 Purple Thought bubbles - Shows reasoning and planning
• ⚡ Blue Action blocks - Displays tool selection and parameters
• 👁️ Green Observation panels - Shows tool execution results
• 💬 Teal Answer sections - Final synthesized responses
• ⚠️ Status indicators - Real-time processing updates

1. Clone and setup the repository
2. Start the backend with repository indexing
3. Launch the frontend for the chat interface
4. Create a new session by entering any GitHub repo URL
5. Watch the AI agent clone, index, and analyze the repository
6. Ask complex questions and watch the agent reason through solutions step-by-step
7. Use @-mentions for specific file context and explore with the file tree
8. Observe real-time reasoning as the agent explores, searches, and synthesizes answers

Create a .env file in the root directory:

1. Visit http://localhost:3000
2. Paste a GitHub repository URL (public repos supported)
3. Wait for repository cloning and indexing (progress shown)
4. Start chatting with the AI about the codebase
5. Use the sidebar to explore files and folders
6. Click any file in the tree to open it in the integrated viewer pane
7. Type @ in the chat for intelligent file/folder autocomplete
8. Mention files with @filename.py or folders with @folder/subfolder

• Overview: "What is this repository about?" - Agent explores structure, reads key files, analyzes patterns
• Architecture: "Explain the project structure" - Agent navigates directories, examines relationships
• Code exploration: "Show me the main components" - Agent searches, categorizes, and explains findings
• Debugging: "Help me understand this error in @file.py" - Agent reads file, searches for related code, analyzes context
• Implementation: "How does the authentication work?" - Agent searches auth patterns, traces code flow

• Thought steps show the agent's reasoning process
• Action steps display which tools the agent chooses to use
• Observation steps show the results of tool execution
• Answer steps provide the final synthesized response

• File references: "Look at @config.py for the settings"
• Folder exploration: "What's in @folder/src/components?"
• Multi-file analysis: "Compare @file1.js and @file2.js"
• Complex queries: "Find all database-related files and explain the schema"
• Autocomplete suggestions: Type @ to see file/folder suggestions with fuzzy search
• Integrated file viewing: Click files in the explorer to view them alongside chat

• Use @ for file autocomplete (optimized for large repos)
• Arrow keys navigate autocomplete suggestions, Enter/Tab to select
• Ask complex questions - the agent will break them down automatically
• Watch the reasoning - each step shows how the AI approaches your problem
• Folder mentions automatically include all files in context
• File viewer pane opens when clicking files in the explorer tree
• The AI remembers conversation history and previous reasoning
• Query complexity automatically adjusts context size (10-25 sources)
• Prompt caching reduces costs for repetitive repository questions

• Understand new codebases quickly with AI-guided exploration
• Debug complex issues across multiple files and components
• Learn best practices from AI analysis of well-structured projects
• Navigate large repositories efficiently with smart search

• Get context on unfamiliar parts of the codebase
• Understand impact of proposed changes
• Generate insights for review comments
• Explore dependencies and related code

• Generate documentation from AI understanding of code
• Create tutorials based on actual implementation
• Explain complex systems with AI assistance
• Understand APIs and integration patterns

• Onboard new developers with AI-guided codebase tours
• Share knowledge through AI-explained code patterns
• Consistent analysis across team members
• Collaborative exploration of system architecture

• Repository cloning with full history and branch support
• FAISS vector indexing for semantic code search with 3000+ documents
• Smart context sizing - automatically adjusts 10-25 sources based on query complexity
• Multi-language support via tree-sitter parsers with intelligent fallbacks
• Query complexity analysis - detects repository overview vs. specific file questions

• OpenRouter prompt caching - 25-90% cost savings on repetitive queries
• Multi-level caching - RAG context, responses, and folder structure
• Async processing - background repository indexing and context retrieval
• Smart file picker - optimized for repositories with 3000+ files (30 folders + 20 files initially)
• Memory management - intelligent conversation context compression

• Session persistence - conversations survive page reloads
• Multiple repository sessions - work with different repos simultaneously
• Real-time status - repository cloning and indexing progress
• Advanced file picker - search, folders, and performance optimization
• @-mention autocomplete - intelligent file/folder suggestions with keyboard navigation
• Integrated file viewer - side-by-side code viewing with syntax highlighting and actions
• Responsive chat interface - mobile-friendly with typing indicators

• Configurable AI providers - OpenAI, OpenRouter, multiple models
• Performance monitoring - cache statistics and usage metrics
• Error handling - graceful fallbacks and detailed logging
• Security - file access restrictions and path validation

• Anthropic Claude: 90% savings on cached content (0.1x cost)
• OpenAI GPT models: 25-50% savings on cached content (0.5x-0.75x cost)
• Automatic caching: Triggers for repository context >1000 tokens

• Index size: 1000-5000+ documents per repository
• Context sources: 10-25 dynamically selected based on query
• Response time: <2s for most queries with caching
• Memory usage: Optimized for large repositories

• Large repos: Handles 3000+ files smoothly
• Initial load: 30 folders + 20 files for fast startup
• Search results: Capped at 100 items for performance
• Tree navigation: Lazy loading for deep directory structures

• Private repository support - GitHub App integration for private repos
• Multi-branch analysis - Compare code across different branches
• Code change analysis - Understand diffs and pull requests
• Integration with IDEs - VS Code extension for seamless workflow
• Team collaboration - Shared sessions and annotations
• Export functionality - Save conversations and insights

We love contributions! Here's how to get started:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

This project is licensed under the MIT License - see the LICENSE file for details.

• Built with React and FastAPI
• UI components from shadcn/ui
• AI providers: OpenAI and OpenRouter
• Vector search powered by FAISS
• Code parsing via tree-sitter