HSCC — Hermes Spark Cluster Control

Turn a DGX Spark GPU cluster into a self-running team of specialized AI agents.

Say “build X” in chat and a fleet of role-specialized agents brainstorms, decomposes, codes, reviews, and lands it — across multiple GPU nodes, hands-off.

HSCC runs on a cluster of DGX Spark (GB10 / Grace-Blackwell) nodes serving LLMs via vLLM, orchestrated by Hermes agents. It is a pure-Python plugin set with three pinned dependencies (PyYAML, rich, psutil — see requirements.txt). Install it with Homebrew (or clone the repo, if you are editing it); hscc bootstrap then copies the plugins into the Hermes runtime (~/.hermes/plugins/), with runtime state in ~/.hscc/.

The design is native-Hermes-first: agent work runs on Hermes’ built-in kanban dispatcher and git worktrees. HSCC does not build a dispatcher — it hooks into the one Hermes already ships, adding the thin physical layer (cluster control, monitoring, model lifecycle), a role framework for specialized workers, and a review gate on the dispatch path. Why this shape (what stock Hermes lacks and HSCC adds):…