Analytics preference

We only load Google Analytics after you opt in. You can change this choice at any time from Cookie settings.

Parallel Execution

Heym runs nodes in parallel when they have no dependencies on each other, and allows multiple workflow runs to execute concurrently.

Heym runs nodes in parallel when they have no dependencies on each other, and allows multiple workflow runs to execute concurrently.

Within a Single Run

DAG-Based Scheduling

The executor groups nodes by execution level using a directed acyclic graph (DAG):

  1. Level 0 – Nodes with no incoming edges (start nodes)
  2. Level N – Nodes whose upstream dependencies are all in levels 0..N-1

Nodes in the same level have no dependencies on each other and run in parallel.

Thread Pool

  • Shared pool: ThreadPoolExecutor(max_workers=8) for node execution
  • Scheduling: wait(..., return_when=FIRST_COMPLETED) – as soon as any node finishes, downstream nodes are scheduled
  • Thread-safe: Node outputs are stored with a lock (store_node_output)

Orchestrator Sub-Agents

When an orchestrator agent returns multiple call_sub_agent tool calls in one turn, they run in parallel. See Agent Architecture.

Merge Node

The Merge node combines outputs from multiple parallel branches. Use it when you explicitly need to combine results; otherwise parallel branches can end in separate output nodes.

Loops

Loop iterations run sequentially. The loop body nodes are re-scheduled per iteration.

Early Return

Output nodes with allowDownstream can return early. Remaining downstream work runs in the background.

Concurrent Workflow Runs

  • No per-workflow lock – Multiple requests can run the same workflow at the same time
  • Async API – FastAPI endpoints use asyncio.to_thread(execute_workflow) so the event loop stays responsive
  • Isolation – Each run gets its own WorkflowExecutor instance and state

Triggers that run workflows (Webhook, Portal, Cron, RabbitMQ, MCP, Editor) each call execute_workflow independently.

Streaming

Streaming uses the same parallel model but emits events as nodes complete. The editor and portal chat UI use streaming internally for real-time progress, and webhook users can now opt into the same event stream through the SSE Streaming execute endpoint.