Pocket Flow

Pocket Flow is a minimalist LLM framework whose entire core fits in roughly 100 lines of Python — a deliberate reaction against bloated agent frameworks. Built by The Pocket team, it gives you just enough structure to express complex agentic workflows as graphs of small, reusable steps, and nothing more. Its motto says it plainly: let agents build agents.

Overview

The framework revolves around three simple ideas: Nodes hold individual steps, Flows wire nodes together by named actions, and shared state carries data between them. On top of that tiny core it supports batch processing, asynchronous execution, and nested flows, which turns out to be enough to build RAG pipelines, multi-agent systems, and self-improving agent loops. Because there is almost no code, you can read the whole framework in minutes, debug it effortlessly, and adapt it to any model provider. Pocket Flow has earned more than 11,200 GitHub stars, is actively maintained with updates in July 2026, and is MIT-licensed. Community ports bring the same philosophy to other languages and runtimes.

Key Features

• Entire framework core is about 100 lines of readable Python
• Graph-based flows compose nodes into sophisticated agent workflows
• Built-in patterns for batch jobs, async execution, and nested flows
• Zero dependencies and no vendor lock-in — works with any LLM API
• Easy to vendor, audit, and modify for production systems
• Growing ecosystem of examples, ports, and agent templates

💳 Pricing

Pocket Flow is free and open source under the MIT license. Your only costs are the model API calls your own workflows make.

Who Is It For

Developers who find mainstream agent frameworks overengineered and want a transparent, hackable foundation for LLM workflows and agents.