ToolsyBio: A retrieval-augmented generation system for navigating the bioinformatics software landscape

Researchers increasingly rely on free and open-source software (FOSS) for computational analysis across the life sciences. However, the growing volume and diversity of available tools make it difficult to discover, understand, and select appropriate software for specific tasks. We present ToolsyBio, a modular system that uses retrieval-augmented generation (RAG) to assist researchers in exploring the bioinformatics software landscape via natural language queries. ToolsyBio is built on structured metadata from the bio.tools registry and semantically enriched with concepts from EDAM, a controlled vocabulary for bioscientific data analysis and data management. The system retrieves relevant tool descriptions using a vector store ChromaDB and generates grounded responses using a locally served large language model (LLM) via Ollama. We describe the system’s architecture, implementation, and potential for improving the findability and usability of bioinformatics tools through a conversational interface.