AI agents · OpenClaw · self-hosting · automation

Quick Answer

Most Popular AI Agent Frameworks for Developers in 2026

Published:

The short answer

LangGraph is the most popular dedicated agent framework for Python developers in October 2026, and the Vercel AI SDK is the most popular for TypeScript. The vendor SDKs — Anthropic’s Claude Agent SDK, the OpenAI Agents SDK and Google ADK — have overtaken the older multi-agent frameworks in downloads, while CrewAI, Agno and Mastra remain the most-starred independent options.

The numbers below were pulled on October 10, 2026 from the PyPI statistics API, the npm downloads API and GitHub.

Python frameworks by downloads

FrameworkPyPI downloads, last 30 daysGitHub starsLicenceBest for
LangGraph44.8M43,000MITStateful, long-running agents with checkpoints and human approval
Strands Agents (AWS)37.1M8,750Apache-2.0Model-driven agents on AWS and Bedrock
Claude Agent SDK29.8M8,200MITClaude Code’s harness (tools, subagents, files) in your own app
OpenAI Agents SDK12.4M29,900MITHandoffs, guardrails and tracing on GPT models
Google ADK10.1M21,800Apache-2.0Gemini agents deployed to Vertex AI Agent Engine
Pydantic AI5.4M20,500MITType-safe agents and structured output
LlamaIndex2.9M52,500MITAgents over your documents (RAG-heavy)
CrewAI2.4M59,500MITRole-based multi-agent “crews”
Agno1.8M42,600Apache-2.0Fast multi-agent apps with built-in memory
smolagents (Hugging Face)0.43M29,800Apache-2.0Code-writing agents, open models
AutoGen0.38M (agentchat)61,300CC-BY-4.0 (repo)Legacy — last push April 15, 2026
Microsoft Agent Framework0.19M14,000MITAutoGen’s successor for .NET and Python

LangChain itself (170.5 million downloads, 147,500 stars) is the most-installed package of all, but most of those installs are for model wrappers and retrieval, not agents; LangGraph is its agent runtime.

TypeScript frameworks by downloads

Frameworknpm downloads, last 30 daysGitHub starsBest for
Vercel AI SDK (ai)124.4M27,200Streaming chat, tool calls and UI in Next.js and Node
Claude Agent SDK52.6M—Claude-based agents and Claude Code automation
LangGraph.js16.2M—Same graph model as Python LangGraph
OpenAI Agents SDK (@openai/agents)8.9M3,900GPT agents with handoffs and voice
Mastra (@mastra/core)8.3M28,700Agents, workflows, memory and evals in one package
Google ADK (@google/adk)1.2M—Gemini agents in TypeScript

How to read these numbers

Downloads overstate infrastructure packages. Strands Agents is installed by AWS tooling and the Claude Agent SDK by Claude Code integrations and CI pipelines, so both rank higher on downloads than on stars. The Vercel AI SDK is a dependency of thousands of Next.js templates.

Stars overstate age. LangChain (2022), AutoGen (2023) and CrewAI (2023) accumulated stars during the first agent wave. AutoGen is the clearest case: 61,300 stars, but no push since April 15, 2026, and Microsoft now points new work to the Microsoft Agent Framework.

Activity is the tiebreaker. Every framework in the tables except AutoGen had a commit in the week before October 10, 2026.

Which one to pick

  • You want the safest Python default: LangGraph. Largest ecosystem, durable execution, LangSmith tracing, and the most tutorials and job-post mentions.
  • You build on one model vendor: that vendor’s SDK — Claude Agent SDK, OpenAI Agents SDK or Google ADK. Less abstraction, first access to new tools such as computer use and hosted MCP.
  • You run on AWS: Strands Agents, which deploys to Bedrock AgentCore.
  • You need typed, validated outputs: Pydantic AI.
  • You want role-based multi-agent teams quickly: CrewAI.
  • You write TypeScript: the Vercel AI SDK for the app layer, Mastra when you need workflows, memory and evals.
  • You are on .NET or Azure: Microsoft Agent Framework, not AutoGen.

For a head-to-head on production features see LangGraph vs CrewAI vs OpenAI SDK vs Claude SDK and, for TypeScript, Mastra vs LangChain vs Vercel AI SDK. Model costs behind any framework are in our current API prices table.

Last verified: October 10, 2026. Download counts are rolling 30-day totals and change daily.

Sources