ArchitectureA supervised JVM-class runtime — OLTP on seven engines, OLAP on three. AI-native, MCP-native, observable as plain SQL.Read the architecture
Está viendo la edición Perú. Está viendo la edición Colombia. You're viewing the Pakistan edition. Cambiar a la edición global →Cambiar a la edición global →Switch to the global edition →

Multi-agent orchestration: parallel sub-agents with timeouts and cancellation

An agent can now delegate to registered sub-agents that run in parallel, each with a per-call timeout, cancellation propagation and a per-instance iteration cap, with agent configuration held in the database.

The agent runtime now supports delegation to sub-agents that execute in parallel. A parent agent dispatches work to registered sub-agents, each described by its purpose so the parent can route to the right one, and the calls run concurrently rather than in sequence. Every delegated call carries a timeout, cancellation propagates from parent to child, and each agent instance has a ceiling on its iteration count.

Orchestration controls

  • Parallel delegation. Sub-agent calls run concurrently; the parent collects their results as they complete instead of blocking on each in turn.
  • Per-call timeout and cancellation. Each delegated call has its own time budget, and cancelling the parent propagates to in-flight children so no sub-agent outlives the request that spawned it.
  • Per-instance iteration cap. Each agent instance carries a maximum iteration count, bounding how long an autonomous loop can run before it must stop.

Configuration

  • Database-driven agent definitions. Agent and skill definitions are configured in the database, so the set of available agents, their descriptions and their tools are managed as data rather than compiled into the runtime.
  • Described sub-agents. Each registered sub-agent carries a description used during prompt construction, giving the parent the context to select among them.

Together these turn single-agent execution into bounded multi-agent orchestration: work fans out across specialised agents while timeouts, cancellation and iteration limits keep the fan-out governed.

See the feature →

← All posts