parallel-tool-calls-integration

Enable parallel tool calls by setting the parallel_tool_calls configuration in your agent core to execute multiple concurrent tool invocations within a single model response. This eliminates sequential bottlenecks and increases agent throughput for independent I/O tasks.

Sequential tool execution bottlenecks occur when traditional agent architectures invoke independent tools one by one, causing high latency. This integration solves it by allowing a single model response to trigger multiple concurrent tool executions for batching and aggregation.

To preserve tool response order during concurrent tool calls, execute tools in parallel while collecting ordered results to comply with API semantics. This ensures tool results return to the model in the required tool_call ordering.

Post-tool middleware injection works with concurrent tool execution by triggering middleware only after the final tool result is collected. This ensures middleware processes the complete aggregated batch of concurrent tool responses.

Resilient error handling continues other tool calls if one parallel execution fails. The system aggregates results with error markers for the failed tools while preserving successful concurrent tool responses, preventing total batch failure.

The parallel tool calls integration applies to OpenClaw builder, harness, and evaluator agents. It modifies these agents to fetch data from several sources or run independent processors concurrently within a single model turn for 2–5x throughput gains.