minion-orchestrator

Background job orchestration for LLM subagent workflows manages long-running, complex reasoning tasks through a Postgres-native queue, ensuring durability, observability, and the ability to steer agents mid-flight.

You can steer a running LLM subagent by pausing, resuming, canceling, or sending directive messages to the active agent, allowing you to change its behavior and reasoning process in real-time.

This orchestration tool supports parallel task execution through parent-child DAGs, enabling fan-out orchestration ideal for large-scale research or data processing workflows with complex dependency management.

You can manage both deterministic shell command execution and complex LLM subagent reasoning loops within a single unified Postgres-backed job queue interface for streamlined background processing.

You need a Postgres database to run durable background jobs, as the orchestration system relies on a Postgres-native queue to satisfy requirements for session persistence and token accounting.