Top AI Agent Frameworks & Harnesses (2026)
An agent framework is the code you build an agent on. It handles the model calls, tool use, memory, and control flow. So you do not write the loop from scratch. It is a different thing from an end-user coding harness like Claude Code or Cursor. This list covers the notable agent-building frameworks and SDKs.
A low-level orchestration framework from LangChain. Use it to build stateful, long-running agents. It has durable execution and human-in-the-loop control.
- License:
- MIT
- Language:
- Python + JS/TS
A standalone Python framework, independent of LangChain. It orchestrates role-playing multi-agent workflows that run on their own.
- License:
- MIT
- Language:
- Python
A Python and .NET framework for building multi-agent AI apps. The community now maintains it, and it is in maintenance mode.
- License:
- MIT (code)
- Language:
- Python + .NET
OpenAI's light framework for building workflows with many agents. It supports many LLM providers.
- License:
- MIT
- Language:
- Python
Google's open-source, code-first Python toolkit, the Agent Development Kit. Use it to build, test, and deploy AI agents. It is not tied to one model or one way to deploy.
- License:
- Apache-2.0
- Language:
- Python
A TypeScript framework for building AI-powered apps and agents. Its core ships under Apache-2.0.
- License:
- Apache-2.0 (core)
- Language:
- TypeScript
A Python agent framework from the Pydantic team. It builds production-grade generative-AI apps. Outputs are type-safe and structured.
- License:
- MIT
- Language:
- Python
Reviewed quarterly. This list is the most durable one here. The main thing that shifts is maintenance status, as a framework can go into maintenance mode. Languages and licenses are stable. Each row carries its last-checked date and source.
Common questions
- What is the difference between an agent framework and a coding harness?
- A framework (LangGraph, CrewAI, the OpenAI Agents SDK) is a library you build your own agent on. A coding harness (Claude Code, Cursor) is a finished product you use to write code. The framework is for building agents. The harness is an agent you use.
- Which agent framework should I pick?
- It depends on your language and how much control you need. LangGraph and Pydantic AI suit typed Python control. Mastra suits TypeScript. The OpenAI Agents SDK is a lightweight multi-provider start. Google ADK is a code-first toolkit not tied to one stack. Check maintenance status too. Some frameworks have moved to maintenance mode.