Nous ResearchHermes Agent

Configure Subagent Delegation — Parallel Task Execution

Set up subagent spawning so Hermes can delegate tasks to child agents for parallel workstreams and complex orchestration.

Subagent delegation lets Hermes spawn child agents for parallel workstreams. Instead of doing everything sequentially, the main agent can delegate tasks to children that work concurrently. Perfect for research tasks, batch processing, or complex multi-step workflows.

Deploy Hermes faster with FlyHermes

Managed cloud · API costs included · Skill library · Cancel anytime

Before you start:

  • Hermes Agent installed
  • Understanding of context isolation
  • Sufficient API quota for multiple parallel agents

Steps

  1. 1

    Understand delegation

    The delegate_task tool spawns child agents with isolated context for parallel work

  2. 2

    Configure max iterations

    Set delegation: max_iterations: 50 to limit tool calls per child agent

  3. 3

    Set concurrency limit

    Set delegation: max_concurrent_children: 3 for parallel child agents

  4. 4

    Enable orchestrator mode (optional)

    Set delegation: orchestrator_enabled: true and max_spawn_depth: 2 for hierarchical delegation

  5. 5

    Override child model (optional)

    Set delegation: model: google/gemini-3-flash-preview to use a different model for children

Pro Tips

  • 💡Child agents have isolated context — they can't see the parent's full conversation history
  • 💡Use batch mode (default 3 parallel) for tasks like 'research these 5 topics'
  • 💡Orchestrator mode (max_spawn_depth: 2+) allows children to spawn their own workers — powerful but costly
  • 💡Override the child model to use a cheaper/faster model for simple delegated tasks

Troubleshooting

Child agents not spawning

Check that delegation: orchestrator_enabled: true is set. Some configs disable delegation by default.

Too many concurrent children causing rate limits

Reduce max_concurrent_children from 3 to 1 or 2 to avoid hitting API rate limits.

Children running too long

Reduce max_iterations from 50 to 20-30 to limit child agent runtime.

High costs from delegation

Set delegation: model: to a cheaper model like gemini-flash for children. Parent uses main model, children use cheap model.

Related Guides