AgentCo-op: Retrieval-Based Multi-Agent Workflow Synthesis
AgentCo-op is a retrieval-based synthesis framework for composing multi-agent workflows in open-ended scientific settings. It assembles reusable skills, tools, and external agents into executable workflows via typed artifact handoffs, applying bounded self-guided local repair when execution fails. In two genomics case studies, AgentCo-op coordinates independently developed agents for spatial transcriptomics and gene-set interpretation without redesign or global topology search. The framework enables collaborative discovery from spatial transcriptomics data and builds parallel workflows for cross-modality tasks.
Key facts
- AgentCo-op is a retrieval-based synthesis framework
- It composes reusable skills, tools, and external agents into executable workflows
- Uses typed artifact handoffs for component integration
- Applies bounded self-guided local repair upon execution failure
- Tested in two open-world genomics case studies
- Coordinates agents for spatial transcriptomics and gene-set interpretation
- Does not require redesign or global topology search
- Enables collaborative discovery from spatial transcriptomics data
Entities
Institutions
- arXiv