Pocket Flow

A TypeScript framework for building LLM-powered applications with: - 🤖 Intelligent Agents - 🧩 Task Decomposition - 📚 Retrieval Augmented Generation (RAG) - And more...

Architecture

We model LLM workflows as a Nested Directed Graph where:

• 🔹 Nodes handle atomic LLM tasks
• 🔗 Actions connect nodes (labeled edges) for agent behavior
• 🔄 Flows orchestrate node graphs for task decomposition
• 📦 Nesting allows flows to be used as nodes
• 📊 Batch processing for data-intensive tasks
• ⚡ Async support for parallel execution

---

Getting Started

• Quick Start
• Setup Steps
• Key Abstractions

Core Concepts

• Agent Basics
• Node Fundamentals
• Understanding Flows
• Data Structures

Features & Extensions

Basic Features

• Async Processing
• Batch Operations
• Communication Patterns
• Task Decomposition

Advanced Capabilities

• Large Language Models
• MapReduce Workflows
• Memory & Caching
• Multi-Agent Systems
• Parallel Execution
• Retrieval Augmented Generation
• Tool Integrations

Additional Resources

• Sample Applications
• Conceptual Paradigms
• Visualizations

---