AI assistant for helping with NeuroML queries and model generation.
Please note that this project is under active development and does not currently provide a stable release/API/ABI.
The package consists of three services:
- the AI assistant backend: this includes the RAG and LLM/agentic logic. This is exposed to users via a FastAPI based REST API.
- the MCP server: this is where we define various "tools" that the LLM can use in its agents.
- the frontend: currently, there's a streamlit UI, but one can use anything to communicate with the REST API---even just the swagger interface.
So, before using the frontend, one must start the MCP server and the FastAPI server.
The devstack.sh script in the scripts folder does this for now.
Please take a look at it and feel free to modify it for your own deployments.
Please note that the ports are currently hard-coded in the code.
The following models are currently being used for testing:
Currently, Ollama is used for local deployments. Please see the code to see what models are currently used. You can modify these to use different models to suit your hardware. However, do note that picking smaller models will most certainly affect the correctness/performance of the RAG. To install Ollama and pull the models, please follow the official documentation: https://ollama.com/download
Since this project uses LangChain, you can use any model that is supported by LangChain. In most cases, you will need to declare some environment variables that will contain your API keys. Please see the LangChain documentation for more information:
https://docs.langchain.com/oss/python/integrations/providers/overview
This code is licensed under the MIT license. Please refer to the licenses of the various LLM models for information on their licensing and usage terms.