Agent Coordination
For tasks that are too large for a single agent turn, OpenHuman can spawn subagents that work in parallel and report back.
How it works
A root agent dispatches work to one or more subagents. Each subagent gets its own context window and tool access. Results are streamed back to the root agent, which synthesizes them.
Tools in the family
| Tool | What it does |
|---|---|
subagent_spawn | Launch a subagent with its own prompt, model, and tools |
subagent_list | See active subagent sessions |
subagent_join | Attach to a subagent's session to inspect or interact |
subagent_cancel | Cancel a running subagent |
Use cases
- Parallel research - one subagent searches the web, another queries the Memory Tree, a third reads a doc
- Codebase-wide refactors - fan out across multiple directories simultaneously
- Review gates - spawn a review subagent while the main agent continues drafting
Tool access scoping
Subagents inherit a restricted toolset by default. The root agent can opt in to additional tools per subagent invocation.