On-site agentic-AI user platform installation - Task 058

Fit to programme

This task has been identified by the working groups as part of the agenda behind WP 2.3.

The task number is 058.

Summary

Agentic AI, where multiple AI LLM agents work together to complete tasks, is becoming key in many researcher and RTP workflows. Unfortunately, the cost of AI subscriptions is high, disadvantaging those who do not have access to funding. Additionally, any data put into queries and instructions then becomes available for training of future AI systems, leading to leaked research data and secrets.

Security, privacy, reliability, and digital sovereignty concerns are pushing many organizations to reconsider their dependence on US-based models and compute infrastructure for code development.

This project will overcome these issues by installing an agentic AI workflow engine on DRI facilities, making use of open models and the hardware already available at these sites to provide agentic AI tooling which is free for researchers to use. These installations will also ensure that no data leaves the site, thus ensuring that researcher data is not available for unintended uses by AI companies.

Methodology

Proposals for this project should first identify an open-source agentic AI platform which is suitable for installation on DRI facilities. The proposals should then identify the intended DRI facility (or facilities) where the tooling will be installed, and the resources required on these facilities. Proposals should describe how installation and testing will be carried out, what resources on the DRI facilities will be used for the inferencing, and how users will be able to access and use the platform. A description of the process used to install the platform and make it ready for users should be make available on the SHAREing website. Proposals should detail the RTP training and upskilling that will be facilitated as part of the work to be hosted on the SHAREing website, and provide documentation in a relevant location about how users can access the platform and make use of it.

Outcomes

The key outputs will be

• a working agent-based programming service, deployed at a DRI facility, which can, in theory, be rolled out to all of the UK
• to enable the latter, an installation journal detailing how the platform was installed on the selected DRI facility, and the upskilling benefits for the RTPs involved, with deliverables being documentation describing this on the SHAREing website. Additionally, the installed platform, made accessible to users will be a key output, and documentation for users detailing how to access it will be a deliverable.