Feature Request: Increase or remove the hard cap on Parallel Block batch size (currently limited to 20)
Current Behavior
The Parallel Block currently enforces a maximum batch size of 20 concurrent executions. When the number of items exceeds this limit, the block automatically falls back to serial batching — processing 20 at a time and waiting for each batch to complete before starting the next.
Problem
This hard cap of 20 is unnecessarily restrictive for use cases that do not involve LLM calls or external API rate limits. For example:
- Processing 50–100 items through pure compute tasks (data transformation, formatting, rule-based routing, math operations)
- Nested Parallel Blocks (Parallel inside Parallel) where the outer block fans out to N branches and each inner block also runs M tasks concurrently — the 20-limit applies at every level independently, causing compounding serial batching even when no rate-limited resources are involved
- Workflows where all parallel tasks are self-contained Function blocks with zero external I/O
In these scenarios, the 20-item cap forces unnecessary sequential waiting, increasing total execution time without any technical justification.
Proposed Solution
Please consider one or more of the following approaches:
- Raise the default cap from 20 to at least 50 or 100
- Allow user-configurable concurrency limit — let users set their own maxConcurrency value per Parallel Block (e.g. 1–200), so they can tune it based on their actual workload and resource constraints
- Distinguish between LLM and non-LLM tasks — apply a conservative default only when the inner blocks contain Agent/LLM nodes, and use a higher (or unlimited) default for pure compute tasks
Use Case Example
With a fixed cap of 20, this degrades into multiple serial batches even though the hardware and platform have no real constraint preventing full concurrency.
Expected Behavior
Users should be able to configure maxConcurrency freely per Parallel Block, or at minimum have a significantly higher ceiling (50–100), so that non-LLM parallel workloads can execute at full speed without artificial bottlenecks.
Why This Matters
Workflow automation platforms derive much of their value from true parallelism. Capping concurrency at 20 — regardless of workload type — undermines this core value proposition for power users who need high-throughput, non-LLM parallel processing.
Thank you for considering this improvement!
Feature Request: Increase or remove the hard cap on Parallel Block batch size (currently limited to 20)
Current Behavior
The Parallel Block currently enforces a maximum batch size of 20 concurrent executions. When the number of items exceeds this limit, the block automatically falls back to serial batching — processing 20 at a time and waiting for each batch to complete before starting the next.
Problem
This hard cap of 20 is unnecessarily restrictive for use cases that do not involve LLM calls or external API rate limits. For example:
In these scenarios, the 20-item cap forces unnecessary sequential waiting, increasing total execution time without any technical justification.
Proposed Solution
Please consider one or more of the following approaches:
Use Case Example
With a fixed cap of 20, this degrades into multiple serial batches even though the hardware and platform have no real constraint preventing full concurrency.
Expected Behavior
Users should be able to configure maxConcurrency freely per Parallel Block, or at minimum have a significantly higher ceiling (50–100), so that non-LLM parallel workloads can execute at full speed without artificial bottlenecks.
Why This Matters
Workflow automation platforms derive much of their value from true parallelism. Capping concurrency at 20 — regardless of workload type — undermines this core value proposition for power users who need high-throughput, non-LLM parallel processing.
Thank you for considering this improvement!