minion-orchestrator

You can orchestrate shell jobs and LLM subagents in parallel using a unified queue system. The minion-orchestrator manages deterministic shell commands and open-ended LLM reasoning tasks, providing observable background execution and fan-out task synthesis.

To run background tasks that survive gateway restarts, use a Postgres-backed durability layer. This approach ensures queued shell jobs and LLM subagent work remain durable and observable across system restarts without losing execution state.

Yes, you need Minions installed and access to the Minions queue to manage background LLM subagent jobs. The orchestrator acts as a unified interface relying on this underlying queue to handle task submission and parallel execution.

You submit a shell job to a background task queue by passing command parameters via CLI, such as `gbrain jobs submit shell --params`. Include your target command and working directory in the JSON payload to execute the deterministic script.

Yes, you can steer running LLM subagents during parallel task execution. The orchestrator allows dynamic steering of active agents, enabling you to adjust open-ended reasoning and tool-using research while background work is in progress.