Introduction to ExtendedLM

ExtendedLM is a comprehensive AI agent platform that provides multi-provider LLM integration, advanced agent systems, visual workflow creation, and RAG capabilities.

Core System Architecture

ExtendedLM operates as a central hub connecting user interfaces, intelligent agents, visual workflows, and various LLM providers.

Figure 1: ExtendedLM Platform Architecture

Core Capabilities

• Multi-Agent System: 12 specialized agents for translation, weather, RAG, PDF reading, computer use, and more
• Unified Gateway: Single API for OpenAI, Anthropic, Google, xAI, Ollama, and local models
• Visual Workflows: Drag-and-drop workflow builder with ReactFlow integration
• Advanced RAG: Vector search, knowledge graphs (GraphRAG), and hybrid retrieval
• MCP Integration: Model Context Protocol support for extensible tools
• Computer Automation: Browser automation, file operations, and shell execution via Mate

Key Features

Agent System

ExtendedLM includes 12 specialized agents, each designed for specific tasks:

Learn more about agents →

Gateway

Unified LLM gateway with OpenAI-compatible API supporting:

• OpenAI (GPT-4o, GPT-5 series)
• Anthropic (Claude Opus 4, Sonnet 4, 3.7 Sonnet)
• Google (Gemini 2.5 Flash/Pro)
• xAI (Grok-4 Fast)
• Ollama (Local models: Phi4, Qwen3, Deepseek-r1, etc.)
• Gateway llama.cpp (Local GGUF models with GPU acceleration)

Learn more about Gateway →

Workflow System

Visual workflow builder with 7 node types (Start, LLM, Tool, Code, Condition, Loop, End) and real-time execution monitoring.

Learn more about Workflows →

RAG System

The ExtendedLM RAG system implements a sophisticated pipeline to ensure accurate and context-aware responses from your document knowledge base.

Figure 2: RAG Ingestion and Retrieval Pipeline

The system utilizes a multi-stage process:

• Ingestion & Chunking: Supports PDF, URL, and text inputs with RAPTOR hierarchical chunking for better context preservation.
• Hybrid Storage: Combines PostgreSQL + pgvector for dense vector storage with a Knowledge Graph for structural relationship mapping.
• Intelligent Retrieval: Uses hybrid search (keyword + vector) with reranking to find the most relevant information.
• Global Caching: Accelerates frequent queries using Valkey Search.

Learn more about RAG →

Quick Start

Get ExtendedLM running in 5 minutes.

1. Clone Repository

# Repository URL will be provided
cd ExtendedLM

2. Install Dependencies

npm install

3. Configure Environment

Copy .env.example to .env.local and configure:

# Supabase
NEXT_PUBLIC_SUPABASE_URL=http://127.0.0.1:54321
NEXT_PUBLIC_SUPABASE_ANON_KEY=your_anon_key
SUPABASE_SERVICE_ROLE_KEY=your_service_role_key

# AI Providers
OPENAI_API_KEY=sk-...
XAI_API_KEY=xai-...
GOOGLE_API_KEY=...

# Valkey/Redis
VALKEY_HOST=127.0.0.1
VALKEY_PORT=6379

4. Start Development Server

npm run dev

Open http://localhost:3000 in your browser.

5. Start Gateway (Optional)

For local model inference:

cd apps/Gateway
export LLAMA_MODEL_DIR=$PWD/../../models/llama
cargo run --release --bin gateway

6. Start Mate (Optional)

For Computer Use agent:

cd apps/Mate
./setup/start.sh

Installation

Prerequisites

• Node.js: 18.x or later
• npm: 9.x or later
• PostgreSQL: 14+ with pgvector extension
• Valkey/Redis: For caching (optional but recommended)
• Docker: For Mate (optional)
• Rust: 1.70+ for Gateway compilation (optional)

Clone Repository

# Repository URL will be provided
cd ExtendedLM

Install Dependencies

npm install

Setup Database

You can use Supabase (cloud or local) or a custom PostgreSQL setup.

Option 1: Local Supabase

# Install Supabase CLI
npm install -g supabase

# Start local Supabase
supabase start

# Apply migrations
supabase db push

Option 2: Supabase Cloud

1. Create a project at supabase.com
2. Run migrations: supabase db push
3. Configure Storage buckets: files, rag-images

Configure Environment

Create .env.local from .env.example:

cp .env.example .env.local

Edit .env.local with your credentials.

Start Development

npm run dev

Access at http://localhost:3000