Skip to main content

​
Process independent items concurrently

Batching runs the same block once for each item in a collection. Use it when items are independent and can be processed concurrently.
Workflow batching in light modeWorkflow batching in dark mode

​
When to use batching

Good candidates include:
  • Extracting the same fields from multiple documents
  • Classifying a list of requests
  • Applying one validation to every record
  • Creating a draft result for each independent item
Use a loop instead when items must run in sequence, share mutable state, or depend on the previous result.

​
Enable batching

1

Provide a collection

Connect an array or table containing the items to process.
2

Select the processing block

Open the block’s settings and enable batching where supported.
3

Configure the item input

Confirm which value from the collection is passed to each block execution.
4

Handle collected output

A batched block returns a collection of results. Connect it only to downstream blocks that can handle that type.
5

Test a small batch

Verify ordering, result shape, and failure behavior before increasing volume.
Example batched workflow in light modeExample batched workflow in dark mode

​
Result handling

Decide how the workflow should behave when one item fails:
  • Stop the full run
  • Return successful and failed items separately
  • Retry eligible failures
  • Send failures for human review
Do not treat a partially successful batch as fully complete unless the downstream process explicitly supports that outcome.

​
Capacity and external systems

Concurrent work can increase load on models and connected applications. Consider:
  • Provider rate limits
  • API quotas
  • File and payload size
  • Duplicate writes
  • Ordering requirements
  • Downstream concurrency limits
  • Usage-based charges
Start with a small representative batch and increase volume only after observing reliable behavior.

​
Limitations

Nested batching is not supported. If a process needs multiple layers of collection handling, simplify the data shape, use a subflow, or process one layer sequentially.

​
Batching checklist

  • Items are independent.
  • Input is an array or compatible collection.
  • Downstream blocks expect a collection.
  • Partial failures have a defined outcome.
  • External rate limits are understood.
  • Duplicate external actions are prevented.
  • Representative volume has been tested.
See Batching tutorial for a guided example.