# HSCC - Hermes Spark Cluster Control

Turns a DGX Spark GPU cluster into a self-running team of role-specialized AI agents built on the Hermes runtime, dispatching coding work across nodes, reviewing it with a dedicated agent, and restarting anything that crashes.

HSCC runs on a cluster of DGX Spark GB10 (Grace-Blackwell) nodes serving large language models through vLLM, orchestrated by Hermes agents. It is a set of pure standard-library Python plugins with no external dependencies, developed in a separate work repository and copied by a bootstrap script into the Hermes runtime's plugin directory, with its own runtime state kept apart from that runtime.

Rather than building its own dispatcher, it hooks into the kanban dispatcher and git worktrees that Hermes already provides, adding cluster control and monitoring, a role framework of more than twenty specialized worker profiles such as architect, coder, reviewer and QA, and a review gate on the dispatch path so changes reach an integration branch only after a dedicated agent checks the diff, tests and specification against it.

Work dispatched to cluster nodes is health-monitored, with crashed models relaunched automatically, and a cron watcher reassigns repeatedly failing tasks to stronger models and notifies the operator only when a task remains stuck.

Website: https://github.com/pom11/hscc