OpenAgents is an open platform for language agents in the wild, built by the XLANG Lab and presented at COLM 2024. Rather than another agent demo, it is a full platform with three working agents and the infrastructure to run, evaluate and extend them with real tools and real websites. The project has more than 4,800 stars on GitHub and is released under the Apache-2.0 license.
The platform ships three reference agents. The Data Agent performs data analysis by writing and executing Python and SQL against real datasets. The Plugins Agent operates over 200 real-world API plugins, selecting and chaining tools to complete tasks. The Web Agent browses and acts on live websites autonomously. Behind them, OpenAgents provides a unified framework covering agent configuration, tool integration, environment interaction and evaluation, with a web interface for interacting with the agents. It was designed as research infrastructure: a place where language agents face messy, real environments instead of sandboxed toy tasks.
The platform logs agent actions step by step, which makes failures inspectable rather than mysterious, and its environments can be extended with new websites, APIs and datasets. Configuration is file-based, so experiments with different models and agent settings can be reproduced and shared.
OpenAgents is free and open source under the Apache-2.0 license. Running it costs only what you spend on the underlying model APIs and your own hosting.
OpenAgents is for agent researchers and builders who want a realistic testbed: a platform where data, plugin and web agents can be studied, benchmarked and extended with their own models and tools.