Concurrency allows multiple tasks to make progress with overlapping execution. Asynchronous tools are useful for external API calls: while one waits for a result, other application code can run. section 2.1.3
In Python, an async method returns a coroutine. Await it from asynchronous code, or use asyncio.run() from synchronous code. An agent can also return an awaitable TaskHandler, allowing submitted tasks to run in the background while other work continues. section 2.1.3 section 4.2.3
These mechanisms support overlapping work at two levels:
- An asynchronous tool call can yield while waiting for an external result. section 2.1.3
- An agent can submit multiple tasks without blocking, and asyncio.gather can execute multiple tool calls in a step. section 4.2.3 section 4.2.3
An asynchronous tool call
A Hailstone tool implemented with AsyncBaseTool is called with await. The same non-blocking approach suits a tool that waits for weather data from an external API. section 2.1.3 section 2.1.3
Figure 2.9 Comparing synchronous and asynchronous tool executions.

Compare blocking synchronous tool execution with non-blocking asynchronous execution. section 2.1.3






























