agentUniverse

agentUniverse is a Python multi-agent framework built on large language models. The README says it grew out of AntGroup's real financial business practice and provides the components for a single agent, collaboration mechanisms between several agents, and a way to inject expert knowledge. Its distinctive feature is a set of "pattern" components. PEER splits work among Plan, Execute, Express and Review agents that iterate on feedback, suggested for event interpretation and industry analysis. DOE uses Data-fining, Opinion-inject and Express agents for data-heavy tasks that need computational precision and expert opinions, such as financial report generation. The README says more patterns are coming.

Install with `pip install agentUniverse`, then follow the guidebook's first-example tutorial. The docs walk from building a single agent (tools, knowledge bases, RAG, configuration, testing, deployment and evaluation) to typical multi-agent applications and reusable agent templates, and a standard project scaffold is provided in the examples folder. Models are plugged in through configuration, and the README lists vendors such as Qwen and DeepSeek among others. A visual agentic workflow platform can also be set up. READMEs are available in English, Chinese and Japanese, and the project is Apache-2.0 licensed.