cd /news/artificial-intelligence/amd-hpc-assistant-open-source-ai-age… · home › topics › artificial-intelligence › article
[ARTICLE · art-171467] src=byteiota.com ↗ pub= topic=artificial-intelligence verified=true sentiment=↑ positive

AMD HPC Assistant: Open-Source AI Agent for Supercomputers

AMD released HPC Assistant, an open-source AI coding agent for supercomputers, on November 17 at SC26 in Chicago, running live on Finland's LUMI supercomputer and deploying to Oak Ridge National Laboratory's Frontier exascale machine. The MIT-licensed tool, hosted at silogen/hpc-assistant on GitHub, lets researchers submit Slurm jobs, profile AMD GPUs with rocprofv3, track per-job energy use, and query cluster-specific documentation in plain English, running the LLM locally and SSHing into the cluster for execution. Pekka Manninen, Director of Science and Technology at CSC – IT Center for Science, said the assistant "understands the full researcher workflow, from job submission to GPU profiling, energy visibility and launching scientific models without manual environment setup.

key takeaway

HPC Assistant runs the LLM on the researcher's local machine and SSHes into the cluster to execute Slurm, rocprofv3 and container operations, using markdown skill files with YAML frontmatter, MCP servers for RAG and cluster tools, and a LiteLLM proxy that supports Claude Code and OpenCode with Aitta

read4 min views9 publishedNov 18, 2026
AMD HPC Assistant: Open-Source AI Agent for Supercomputers
Image: Byteiota (auto-discovered)

AMD shipped an open-source AI coding agent for supercomputers on November 17 at SC26 in Chicago. Called HPC Assistant, it runs live on LUMI — Finland’s flagship supercomputer — and is deploying to Frontier, the world’s first exascale machine at Oak Ridge National Laboratory. Researchers can now submit Slurm jobs, profile AMD GPUs, and query cluster documentation in plain English. The sysadmin ticket queue gets shorter.

HPC Is Hard. That’s the Problem This Solves. #

Using a modern supercomputer is not like using a cloud VM. You’re writing Slurm job scripts with the right partition names, memory limits, and GPU counts. You’re managing module environments where conflicting library versions silently break your code. You’re reading system-specific documentation that’s often 200+ pages long and updated quarterly. Researchers without a CS background — the majority of HPC users — routinely spend weeks on this overhead before writing a single line of science code. HPC Assistant’s bet is that an AI agent with access to cluster-specific documentation can absorb that operational burden. Not by giving you better docs to read, but by handling the operations directly.

What It Does (Specifically) #

The tool ships as a set of site-specific skills covering the most time-consuming parts of cluster work:

  • Slurm job management — Submit, monitor, and cancel jobs viasbatch ,srun ,squeue , andscancel commands, issued in natural language
  • AMD GPU profiling — HIP/ROCm application profiling withrocprofv3 : kernel tracing, performance counters, cache efficiency metrics
  • Energy monitoring — Per-job energy consumption tracking via Omnistat and Grafana dashboards
  • RAG-powered documentation — Retrieves answers from LUMI’s or Frontier’s actual documentation: storage policies, project quotas, partition requirements — your cluster’s answers, not generic ones
  • Scientific agent launching — Spin up domain agents like OpenFold3 for protein structure prediction without manually configuring environments

“The HPC Assistant brings something genuinely new to LUMI. It understands the full researcher workflow, from job submission to GPU profiling, energy visibility and launching scientific models without manual environment setup.”

Pekka Manninen, Director of Science and Technology, CSC – IT Center for Science

How It Works #

The architecture explains why this works at all. HPC clusters don’t allow you to run large language models on login nodes — resource limits prevent it. So AMD HPC Assistant uses a local execution model: you run the agent on your own machine, and it SSHes into the cluster to execute operations. The LLM runs locally (or via API); the cluster work happens remotely.

Under the hood, the stack has four layers: markdown skill files with YAML frontmatter define what operations the agent can run; MCP servers expose the RAG knowledge base and cluster tools; a LiteLLM proxy translates between Anthropic and OpenAI-compatible backends; and Singularity/Apptainer containers handle cluster-side execution without system-wide installs.

It supports Claude Code and OpenCode as agent frameworks. For backends, you can use an Aitta token (LUMI-native), an Anthropic API key, or any OpenAI-compatible endpoint.

Open Source, Self-Hostable #

The repo is silogen/hpc-assistant on GitHub, MIT licensed for AMD-authored components. Any HPC center can deploy their own instance and load it with their own site-specific knowledge base and skills. That’s the key differentiator from asking ChatGPT about Slurm: HPC Assistant’s answers are grounded in your cluster’s actual configuration, not generic Linux documentation.

Frontier support is in progress. Dr. Antigoni Georgiadou from Oak Ridge National Laboratory confirmed AMD is working to bring HPC Assistant to Frontier, though no rollout date has been announced.

AMD’s Broader SC26 Bet #

HPC Assistant sits inside a larger AMD push at SC26. The company announced the MI430X GPU — its first chip specifically designed for the convergence of high-precision FP64 simulation and AI, with 432 GB of HBM4. The new Lux supercomputer (No. 14 on TOP500) built by HPE with MI355X GPUs delivers 220.79 petaflops and is designated as the first U.S. AI Factory for the Department of Energy’s Genesis Mission. AMD now powers 202 of the 500 fastest supercomputers — over 40% of the list.

The pattern is deliberate: build the hardware, build open-source software tools, get centers to adopt them, sell more hardware. HPC Assistant is as much a go-to-market play as it is a developer tool — which is not a critique, just context for evaluating AMD’s long-term commitment to maintaining it.

Getting Started #

If you’re on LUMI today, HPC Assistant is live. For other clusters, the GitHub repo covers local mode setup: configure SSH multiplexing, run hpc-assistant-local-setup with your remote host credentials, and launch your agent framework. The skill files are readable markdown — adding site-specific knowledge for a new cluster is a documentation exercise, not an engineering project. The full AMD SC26 announcement and LUMI’s SC26 highlights have additional context on the broader deployment.

AI agents have been colonizing the developer stack from the top down: code editors, then cloud infrastructure, now HPC clusters. If you run workloads on supercomputers, this is worth fifteen minutes of your time.

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @amd 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/amd-hpc-assistant-op…] indexed:0 read:4min 2026-11-18 · —