agentUniverse

Why we included this project

agentUniverse is a Python framework for teams that want several LLM agents cooperating on one complex job rather than running separate prompts. Install it via pip and you get reusable collaboration patterns: the PEER sequence (Plan, Execute, Express, Review) for breaking down reasoning tasks, and the DOE trio for data-heavy work such as financial report generation that needs precision. It grew out of Ant Group's own finance business, so the bundled patterns read like they were tested against real operating constraints instead of written as textbook examples. Each agent's underlying model is set in a simple TOML file, with providers including Qwen, DeepSeek, OpenAI, Claude, and Gemini among others. If you are comparing multi-agent coordination approaches, this gives you a concrete structure to experiment with, plus a visual workflow tool and MCP server integration to speed up prototyping.