Agentic Chat

Multi-modal AI chat platform with intelligent query routing, semantic caching, RAG-powered document intelligence, and autonomous research capabilities.

Overview

Conversational AI system built with Next.js 15, React 19, and PostgreSQL. Features dynamic query classification, specialized tool routing, multi-agent orchestration, and comprehensive document processing with OCR support.

📖 Setup Guide - Installation and configuration instructions
📹 Feature Demos - Video walkthroughs of all major features

Core Features

🔐 Authentication & Security

• Google OAuth Integration via Better Auth
• BYOK (Bring Your Own Key) - AES-256-GCM encrypted API key storage
• Server-side API Proxy - Eliminates client-side key exposure
• Rate Limiting & Usage Tracking - Built-in cost management

🤖 Multi-Model Support

• GPT-5 Series: GPT-5, GPT-5 Mini, GPT-5 Nano (with extended reasoning)
• GPT-4.1 Series: GPT-4.1, GPT-4.1 Mini, GPT-4.1 Nano
• GPT-4o Series: GPT-4o, GPT-4o Mini
• Vision Capabilities: All models support multimodal input (text + images)
• Context windows up to 1M+ tokens

🛠️ Specialized Tools

Web Search

• Real-time internet search via Tavily API
• Intelligent multi-search planning in advanced mode
• Search depth modes: Basic (6 results) and Advanced (10-25 results with AI planning)
• Image results with gallery view
• Source deduplication

YouTube Analysis

• Global Video Caching - PostgreSQL-backed caching with automatic expiration and access tracking
• Multi-tier Transcript Fallback - Tries youtube-transcript, yt-dlp service, then official API
• Transcript extraction with multi-language support
• Automatic chapter parsing from video descriptions
• Video search and batch processing
• Metadata enrichment via YouTube Data API
• Cache statistics and performance monitoring

URL Scraping & Context Injection

• Automatic URL detection and content extraction (up to 5 URLs)
• Smart content extraction with Readability + Cheerio
• Rich metadata and link previews with Open Graph
• 1-hour content caching for performance

Google Suite Integration

• Gmail: Search, read, send, reply, delete emails, modify labels, get attachments
• Google Calendar: List, create, update, delete events with attendees
• Google Drive: Search, list folders, read/create/delete files, move, copy, share
• Google Docs: Create, read, append, find-and-replace
• Google Sheets: Create, read, write, append, clear ranges
• Google Slides: Create, read, add slides
• OAuth 2.0 authentication with automatic token refresh

Deep Research

• Multi-step autonomous research with usage limits (3/month)
• LangGraph-powered workflow: gate check → planning → parallel research → aggregation → evaluation → formatting
• Intelligent gate system to determine research necessity
• Iterative quality evaluation with adjustable strictness

📚 Document Intelligence (RAG)

• Multi-format Support: PDF, DOCX, DOC, TXT, XLS, XLSX, CSV
• Image Processing: JPG, PNG, GIF, WEBP, BMP, SVG, TIFF, ICO
• Semantic Search - pgvector-powered embeddings
• Re-ranking - Cohere reranker for improved relevance
• Chunking & Tokenization - Optimized text splitting

🧠 Memory & Context

• Semantic Caching - Response caching based on query similarity with pgvector
• Performance-optimized Cache Actions - Server actions for efficient cache operations
• Conversation Memory - Cross-conversation context retention via mem0
• Token Usage Tracking - Real-time token counting with context limit notifications
• Message Versioning & Branching - Tree-based conversation versioning with sibling management
• Intelligent Context Router - Dynamic routing with 5 modes:
  • MemoryOnly: Uses conversation memory
  • DocumentsOnly: RAG-based document retrieval
  • VisionOnly: Image-only analysis
  • Hybrid: Combined vision + document context
  • ToolOnly: Direct tool execution without context
• Smart Query Detection - Pattern-based detection of memory, referential, and standalone queries
• URL Context Injection - Automatically scrapes and includes web page content

📤 Export & Sharing

• Export Formats: JSON, Markdown, PDF
• Customizable Export Options: Include/exclude attachments, versions, metadata
• Public Sharing - Share conversations via unique links with OpenGraph previews
• Conversation Management - Rename, delete, organize
• Version History Export - Include edit history in exports

💬 Chat Features

• Real-time Streaming - Progressive response generation with abort support
• Continue Conversation - Automatically resume incomplete conversations with context preservation
• Context Limit Warnings - Visual notifications when approaching token limits with usage statistics
• File Attachments - Up to 5 files per message (documents + images)
• Clipboard Image Paste - Direct image paste from clipboard with auto-naming
• Processing Status Tracking - Real-time document processing progress with chunk/token counts
• Citations Display - Source tracking for RAG responses with relevance scores
• Search Image Gallery - Grid display of search images with expandable view and mobile optimization
• Follow-up Questions - AI-generated conversation prompts
• Message Actions - Edit, regenerate, copy, delete after point
• Version Navigation - Navigate between message versions with visual indicators
• Image Lightbox - Full-screen image viewing with keyboard shortcuts
• Dark/Light Theme - System-aware theme switching with custom logo support
• Tool Selection UI - Visual tool picker with gradient indicators
• Link Previews - Rich metadata cards for URLs in messages

Tech Stack

Frontend

• Next.js 15 - React framework with App Router
• React 19 - Latest React features
• TanStack Query - Data fetching and caching
• Tailwind CSS 4 - Utility-first styling
• Radix UI - Accessible components
• Framer Motion - Animations
• Shiki - Syntax highlighting

Backend

• PostgreSQL - Primary database
• Prisma - ORM and migrations
• pgvector - Vector similarity search
• Better Auth - Authentication
• UploadThing - File uploads

AI & ML

• OpenAI API - LLM inference with vision support
• LangChain - Agent orchestration
• LangGraph - Multi-agent workflows
• LangSmith - LLM observability and tracing
• Cohere - Reranking
• Tavily - Web search
• mem0 - Conversation memory
• Google APIs - YouTube Data API & Google Workspace APIs

Document Processing

• pdf-parse - PDF text extraction
• mammoth - DOCX processing
• word-extractor - DOC files
• xlsx - Spreadsheet parsing
• youtube-transcript - YouTube caption extraction
• @mozilla/readability - Article content extraction
• cheerio - HTML parsing and manipulation

Architecture

Intelligent Context Router
Analyzes queries for images, documents, URLs, and intent to determine optimal routing strategy (Memory, Documents, Vision, Hybrid, or Tool-specific) with pattern-based detection.

URL Scraping Pipeline
Detects URLs in messages, extracts content using Readability/Cheerio, caches for 1 hour, and injects as context. Supports Open Graph link previews.

Intelligent Search Planning
Advanced mode uses LLM planning to analyze query complexity, decompose into targeted sub-queries, execute parallel searches, and synthesize results.

Semantic Caching Layer
pgvector-based similarity caching reduces redundant API calls for semantically similar queries.

RAG Pipeline
Documents processed through format-specific loaders, chunked, embedded, stored in pgvector, and retrieved with semantic search + Cohere reranking. Real-time status tracking and batch operations supported.

Multi-Agent Orchestration
LangGraph-powered workflows with access to RAG, web search, YouTube, Google Suite, and URL scraping tools. Deep research uses multi-phase workflow with quality evaluation (3/month limit).

Vision Processing
Multimodal support across all models for image analysis, OCR, and hybrid vision+document workflows. Search results include image galleries.

Google Suite Integration
OAuth 2.0 access to Gmail, Calendar, Drive, Docs, Sheets, and Slides with automatic token refresh.

YouTube Processing
Global caching system stores video metadata, transcripts, and analysis in PostgreSQL. Multi-tier fallback tries youtube-transcript library, then yt-dlp service, then official YouTube API. Caches expire based on access patterns with built-in statistics tracking.

Message Versioning System
Tree-based message storage with parent-child relationships and sibling indexing enables conversation branching, version navigation, and soft deletion.

Server-side API Proxy
All OpenAI API calls route through Next.js API routes, keeping user API keys encrypted (AES-256-GCM) and enabling server-side rate limiting.