Skip to content
Get packs

PocketFlow

A tiny Python framework (about 100 lines) for wiring LLM calls into graphs of steps and agents.

GitHub Download source
Stars
11.3k
Forks
1.2k
Licence
MIT
Language
Python
Last update
3 months ago
Type
Library

What it is

Pocket Flow is a minimalist framework for building LLM applications, based on a graph abstraction where small nodes are connected into flows. The README says the core is about 100 lines with no dependencies and no vendor-specific wrappers, and that it can be used to build agents, workflows, and retrieval-augmented generation (RAG) systems. People use it when they want to understand and control their LLM code instead of relying on a large framework.

  • Build a basic chat bot that keeps conversation history (cookbook example)
  • Create a research agent that searches the web and answers questions (cookbook example)
  • Run a batch job that translates markdown into multiple languages (cookbook example)
  • Build a retrieval-augmented generation process (cookbook example)

Great for

  • Developers who want a small, readable LLM framework they can inspect in full
  • Python programmers building agents, multi-step workflows, or RAG pipelines
  • People who want to copy the source directly into a project instead of adding a dependency
  • Developers experimenting with AI coding assistants that generate agent code

Not ideal for

  • Teams that need many prebuilt integrations with vendors and vector databases, since the README says it includes none ('None' in its comparison table)
  • Non-programmers, because it is a code library rather than a ready-to-use app
  • Projects needing a formal, versioned release, since the repository shows no latest release
Step-by-step guide: how to use PocketFlowInstall, first run, examples, tips and FAQs
PocketFlow alternativesLangChain vs PocketFlowLangGraph vs PocketFlowCrewAI vs PocketFlow

How to get it

  1. 1

    Install with a package manager

    You'll need: Python

    Install from PyPI

    pip install pocketflow
  2. 2

    Run it

    • Install the package with pip install pocketflow, or copy the source file pocketflow/__init__.py into your project.
    • Open the basic tutorials listed in the README, starting with the Chat example.
    • Read the documentation at the-pocket.github.io/PocketFlow for how nodes and flows are defined.
    • Run one cookbook example and adapt it to your own task.
  3. 3

    Get the most out of it

    • Start with the cookbook examples in order of difficulty, from Chat to Multi-Agent
    • Use the Discord community linked in the README to ask questions about building with it
    • Try the sibling implementations in TypeScript, Java, C++, Go, Rust, or PHP if your stack is not Python
    • Follow the design pattern guides for agents, workflows, and RAG in the documentation

Downloads

Source code (.zip)main branch · straight from GitHubOfficial website & docsthe-pocket.github.io/PocketFlow/

Strengths

  • Very small codebase (about 100 lines), so the whole framework is easy to read and audit
  • No dependencies and no vendor lock-in, according to the README
  • Includes tutorials from a dummy-level chat example up to multi-agent and parallel demos

Trade-offs

  • Provides few built-in integrations, so you write your own connections to LLM providers and tools
  • The README is truncated in the source provided, so some details could not be verified
  • Documentation and examples are spread across the README, a docs site, and a cookbook folder

How it compares

ProjectStarsTypeLevelUpdatedBest for
PocketFlow (this)11.3kLibraryintermediate3 months agoDevelopers who want a small, readable LLM framework they can inspect in full
langchain149kFrameworkintermediate4 days agoPython developers building LLM-powered applications
langgraph44.3kLibraryintermediate4 days agoPython developers building multi-step AI agents that must survive failures or long waits
crewAI60.4kLibraryintermediate4 days agoPython developers prototyping multi-agent workflows
ruflo80.6kDev toolintermediate4 days agoDevelopers already using Claude Code or Codex who want multiple agents working together
PocketFlow-Tutorial-Codebase-Knowledge12.7kDev toolintermediate5 months agoDevelopers who need to understand a large unfamiliar open-source project

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.