AskALCF: An AI Support Assistant¶

Disclaimer

AskALCF is an AI-assisted tool designed to help ALCF users quickly access information. Responses are generated using automated methods and may not always be accurate, complete, or current. Users should confirm critical information through official ALCF documentation or by contacting ALCF support.

Feedback

We value your input to improve AskALCF. If you encounter incorrect or unclear information, please share your feedback to help us enhance the service.

AskALCF is an intelligent AI support assistant designed to enhance user support at the Argonne Leadership Computing Facility (ALCF). It combines two core capabilities:

Retrieval-Augmented Generation (RAG): Provides accurate, context-aware responses by retrieving relevant information from a curated HPC-specific knowledge base before generating answers.
HPC Operation MCP (Model Context Protocol): Enables real-time queries of machine and job status for ALCF systems — Aurora, Polaris, Sophia, and Crux — by interfacing with runtime machine activity data and the ALCF Facility API.

AskALCF Architecture — AskALCF architecture: Orchestrator LLM routes each user query to the RAG pipeline, the HPC Operation MCP, or both.

Knowledge Base¶

Below are the documents curated in the vector database used by AskALCF.

Data	Type / format
ALCF User Guide	Markdown
OLCF User Guide	RST
NERSC Documentation	Web
LLNL User Guide	Web
PBS Documentation	PDF
Slurm Documentation	Web
CUDA C Programming Guide	Web
SYCL Documentation	PDF
AMD HIP Documentation	Web
Intel OneAPI Documentation	Web

Accessing AskALCF¶

The chatbot is currently hosted on a CELS Virtual Machine accessible through https://ask.alcf.anl.gov

Example Questions¶

Knowledge Base Queries¶

Ask specific questions about ALCF systems, programming models, and HPC documentation. The more specific your question, the better the response. Examples include:

How do I log in to Aurora?
How do I request 8 Aurora nodes all within the same rack using PBS, using the following job parameters? Script:pbs_submit_script.sh, Project: RackTest, Queue: prod, Duration: 1 hr, Filesystem: grand.
Will my ALCF HPSS home directory remain accessible for my group members after my account termination?
What is the best environment variable setting for oneCCL?
How should I use copper to run a Python job at scale on Aurora?
What is the optimal CPU binding on Aurora?

Example: Knowledge base query — Example: Answering a knowledge base query with references

Machine & Job Status Queries¶

AskALCF can retrieve real-time machine and job status for ALCF systems (Aurora, Polaris, Sophia, Crux) via the HPC Operation MCP. Examples include:

How many jobs are running on Sophia?
Which machine is most available right now?
What is the current status of Aurora?

Example: Job status query — Example: Querying job status on Sophia via HPC Operation MCP

Example: Machine availability query — Example: Comparing machine availability across ALCF systems

Providing Feedback¶

If you have feedback on a specific response, you can rate and comment directly:
- Click on the stars under "Rate the Answer" to submit a rating
- Click "Add Feedback Comment" to provide additional comments
For high-level comments or feedback, please contact: support@alcf.anl.gov
All feedback is anonymous by default. If you would like to receive follow-up communication, please provide your contact info.