Welcome to Klea¶

Knowledge vaLidated Expert AI Assistant for Neuroscience.

Klea is a suite of AI tools for Neuroscience. It provides a general purpose agent with coding capabilities, a generic RAG pipeline, and MCP servers for modelling and analysis.

Architecture¶

The project is organised as a monorepo with four installable packages:

Directory	Package	CLI	Purpose
`utils_pkg`	`klea_utils`	`klea-stores-create`	Shared utilities, vector store management, base graph classes
`rag_pkg`	`klea_rag`	`klea-rag`, `klea-rag-serve`	Generic RAG pipeline with multi-domain support
`agent_pkg`	`klea_agent`	`klea`, `klea-serve`	General purpose agent with coding capabilities
`mcp_pkg`	`neuroml_mcp`	`nml-mcp`	MCP server for NeuroML tooling

Each package is built on a shared foundation in klea_utils, which provides configurable LLM setup (with runtime model switching), vector store abstraction (Chroma / PGVector / Qdrant), and the BaseLangGraph orchestrator framework. Web interfaces are available via NiceGUI (primary) and Streamlit.

klea_agent is the main application: a general purpose agent that can also code. klea_rag is an additional component, primarily consumed by klea_agent as a retrieval/RAG service.

Prototype Deployments¶

These prototype Klea RAG deployments are available on HuggingFace that use the web interface.

Please note that there are limited resources/credits available for these prototypes, and so they may fall over if there is too much activity. They are not production deployments.

Funding¶

Klea is funded by the BioFAIR Pathfinder Projects grant “Creating AI-enabled analysis pipelines for FAIR neuroscience data”, awarded to Padraig Gleeson and Ankur Sinha at University College London.