Try CUGA in Hugging Face, the #1 Generalist Agent in the AppWorld Leaderboard (huggingface.co)

IBM has launched CUGA (Configurable Generalist Agent), an open-source AI agent designed to automate complex, multi-step tasks across web and API environments. CUGA stands out in the AI/ML community by addressing challenges faced by existing frameworks, such as brittleness and tool misuse, while offering high performance on industry benchmarks. It currently ranks #1 on the AppWorld benchmark, demonstrating its robustness with 750 real-world tasks, and has shown strong capabilities in the WebArena for autonomous web agents. CUGA's architecture allows developers to abstract orchestration complexity, enabling them to focus on domain-specific requirements rather than the intricacies of agent construction. Key features include configurable reasoning modes for optimizing performance and cost, seamless multi-tool integration, and the ability to compose tasks using specialized agents. Furthermore, its integration with Hugging Face Spaces simplifies experimentation with open models, making CUGA an accessible solution for both developers and enterprises. By embracing a fully open-source approach under the Apache 2.0 license, CUGA not only promotes flexibility in AI agent building but also encourages community collaboration and innovation.