OpenAgents

What people say

The newest entry in OpenAgents' own changelog is from October 2023, and that tells you most of what you need to know. The XLang Lab project — built at the University of Hong Kong with Sea AI Lab — was a genuinely early, genuinely open attempt at productised agents: a Data agent that writes and runs Python and SQL against your uploads, a Plugins agent with 200+ tool integrations, and a Web agent driving a Chrome extension. The accompanying paper landed at COLM 2024, and the repo has collected around 4.8k GitHub stars.

As a product, though, it's frozen in the ChatGPT-plugins era it was built to answer. Deployment means standing up a Flask backend, a React frontend and your own model API keys — the low-setup pitch was relative to research frameworks of 2023, not to anything current. Open issues and pull requests sit largely unattended, and the hosted demo at chat.xlang.ai is best treated as intermittently available. Meanwhile the agent-framework field it helped open has been lapped by faster-moving stacks — LangGraph, CrewAI, and the vendor agent SDKs — that ship weekly.

It still earns a place as a reference. The codebase is readable, the paper is a solid systems write-up of what breaks when non-expert users meet agents, and for a student or researcher dissecting agent architecture there are worse teachers. Just don't mistake it for a maintained platform. Nothing here is production infrastructure, and treating it as a free Code Interpreter substitute in 2026 will mostly teach you about dependency rot.

Summary of public user & expert reviews, compiled by RECATOOLS.