An AI coding assistant built for high-performance computers
Built for scientists, engineers, and data teams working on HPC systems advanced by AMD, the HPC Coding Assistant is an AI-powered tool that runs directly on a supercomputer's login node. It combines a general-purpose coding agent like Claude Code or OpenCode with HPC-specific knowledge, site documentation, job scheduling know-how, and GPU profiling guidance. The result is fast answers and smooth task execution, without the need to constantly search for information.
Everything runs inside containers, requiring no installation on the shared cluster.
Key Features
What you can do with the HPC Coding Assistant
By translating plain instructions into scripts and containers, managing resources, explaining errors, and optimizing performance, the HPC Coding Assistant helps teams achieve quick initial results, simple scaling, low support requests, and efficient use of computing resources.
Ask questions about the system.
Query quotas, partitions, storage limits, and more, with responses grounded in your site's actual documentation, not AI guesswork.
Write and manage compute jobs.
Assistance on drafting job scripts, picking the right resources, and troubleshooting failures.
Profile GPU workloads.
Walk through AMD GPU profiling with step-by-step assistance.
Accelerate general coding tasks.
Query quotas, partitions, storage limits, and more, with responses grounded in your site's actual documentation, not AI guesswork.
How it Works
1
Build containers.
Use any machine with Docker installed (such as your laptop), then transfer them to the cluster.
2
Run setup wizard on login node.
The wizard guides you through selecting your site, AI backend, and preferred agent, then handles the rest.
3
Launch with a single script.
A single script handles day-to-day startup and keeps services healthy across login-node changes.
Current Version Limitations
Currently supports LUMI only
The skills and documentation are tailored for the LUMI supercomputer. The framework is designed to support additional sites, but they are not yet configured.
AI backend required
Either an Aitta token (on LUMI) or an Anthropic API key with outbound network access.
Documentation is a snapshot
The knowledge base is built at setup time and won't pick up upstream changes until you rebuild it.
Best-effort safeguards
The assistant is instructed to avoid destructive actions and request confirmation before submitting jobs. This guidance is prompt-level and does not constitute a hard security boundary.
