hpcGPT is a shared CLI foundation built on top of the OpenCode agent for HPC centers. Each center keeps its own deployment code, configuration, prompts, MCP servers, and site-install artifacts in its own top-level directory.

Current Deployments

• NCSA/ — NCSA deployment for Delta (support assistant, MCP servers, site module install)

TL;DR - Getting Started

On a deployed cluster (end users):

module load hpc-gpt/1.15.13
opencode

Local development:

curl -fsSL https://opencode.ai/install | bash
export OPENCODE_CONFIG=/absolute/path/to/this/repo/NCSA/client-deployment/opencode.jsonc
export NCSA_LLM_URL=https://your-endpoint/v1
opencode

Pick the center deployment you want to run by setting OPENCODE_CONFIG to that center's config file. See the center's README for environment variables, MCP setup, and site-admin instructions.

Repository Layout

hpcgpt-cli/
  README.md
  LICENSE
  favicon.png
  NCSA/
    README.md
    client-deployment/     # Site install: installer, modulefile, config, prompts
    mcp_servers/           # Slurm, Illinois Chat, report, ticket knowledge-base MCP servers
    ticket-ingest/         # Support ticket → Q&A dataset pipeline
    doc-scraping/
    example.env

Contribution Model (Per Center)

When adding support for another university or supercomputing center, create a new top-level directory (for example CenterName/) and keep center-specific content scoped there.

Required in each center folder

• README.md describing architecture, tools, setup, and operations
• client-deployment/ (or equivalent) with site-install artifacts: config, prompts, installer, and modulefile templates
• mcp_servers/ for MCP servers owned by that center
• example.env and optional additional env examples
• OpenCode config (opencode.jsonc) with that center's providers, agents, models, and MCP wiring

Guidelines

• Keep secrets out of git; commit only example env files.
• Keep center-specific naming and endpoints inside that center folder.
• Keep root-level docs and files generic and reusable across centers.
• Update this root README.md when adding a new center directory.
• Prefer shared patterns, but allow center-specific implementation details.

Shared Expectations

• Deployments can use OpenAI-compatible providers; exact provider configuration is center-specific.
• MCP servers and permissions should be documented in each center's README.md.
• Site-admin install runbooks belong in each center's client-deployment/ (or equivalent) directory.
• Operational runbooks, support flows, and escalation contacts belong in each center folder.

NCSA Deployment

For architecture, MCP servers, development setup, and Delta site-admin instructions, see NCSA/README.md.

Site-admin install steps: NCSA/client-deployment/README.md.

License

MIT — see LICENSE.