golem-parallel-workers-ts

You can implement parallel agent workers by spawning multiple Golem agent instances concurrently using Promise.all, then collecting and aggregating all returned outputs into a unified result set for fan-out fan-in workflows.

The best way to handle errors in concurrent child agents is using Promise.allSettled for error isolation. This allows the parent workflow to collect results from all spawned workers without failing entirely if one child agent encounters an error.

Fork-based parallelism works by cloning the current agent at its current execution point to run additional logic concurrently. This allows multiple branches to execute in parallel while maintaining access to the shared state of the original agent.

Yes, you can use Golem promises to manage long-running concurrent workflows by dispatching work with PromiseId-based trigger and await patterns. This provides durable async fan-out with reliable result collection and correct synchronization to avoid deadlocks.

You can limit concurrency load through chunked concurrency control, which batches multiple spawned workers to restrict active instances while still achieving parallel throughput and aggregating outputs efficiently.