Build Durable Agents With Pydantic AI And Temporal

Pydantic AI has introduced a native integration with Temporal, a durable execution engine, enabling agentic applications to resume from failures rather than restarting. The TemporalAgent class wraps existing agents, converting tool calls and model requests into Temporal Activities that can be retried on failure. This integration allows workflows to be replayed from event history, ensuring progress is preserved across crashes.

Agentic applications can fail mid-run: a network call drops, a server process crashes, an LLM request times out. Without durability built in, these failures mean lost progress and repeated wasted LLM calls when the process has to start over. Pydantic AI provides a native integration with Temporal , a durable execution engine. Temporal records each step of a workflow's execution as an event history, so if a process crashes, it can resume from where it left off instead of starting over. Pydantic AI exposes this through the TemporalAgent class, which wraps an existing agent and converts its tool calls, MCP calls, and model requests into Temporal Activities. python from pydantic ai import Agent from pydantic ai.durable exec.temporal import TemporalAgent planning agent = Agent model='anthropic:claude-fable-5', temporal agent = TemporalAgent wrapped=planning agent You call .run ... on temporal agent the same way you would on the underlying agent. The difference is under the hood: each tool call, MCP call, or model request now runs as a Temporal Activity. Activity — a function that performs a unit of work: an LLM call, an API request, a file write. If it fails, Temporal can retry it based on a configured retry policy. python from temporalio import activity @activity.defn async def create sandbox user id: str - str: sandbox = await AsyncSandbox.create template="node-expo-builder" return sandbox.sandbox id Workflow — orchestrates Activities, defining what runs and in what order. Workflow code must be deterministic no direct network calls, no random values, no reading the system clock , since Temporal replays it to reconstruct state. python from temporalio import workflow from datetime import timedelta @workflow.defn class ExampleWorkflow: @workflow.run async def run self, prompt: str - str: sandbox id = await workflow.execute activity create sandbox, args= self.user id , start to close timeout=timedelta minutes=2 , return sandbox id Worker — polls a task queue for scheduled work and executes it. A Workflow or Activity must be registered with a Worker before it can run. python import asyncio from temporalio.client import Client from temporalio.worker import Worker async def main : client = await Client.connect "localhost:7233" worker = Worker client, task queue="coding-agent", workflows= ExampleWorkflow , activities= create sandbox , await worker.run if name == " main ": asyncio.run main The task queue name is the only link between the code that starts a Workflow and the Worker that executes it — they never call each other directly. Workers run Activities and Workflows differently: Activities run their function body when scheduled, and may be retried on failure per their retry policy. Workflows are replayed against their event history each time they need to advance, which is what lets a Workflow resume correctly on a new Worker after a crash. Once a Workflow is running, Signal and Query let you interact with it from outside. Signal sends data into a running Workflow, asynchronously, with no return value: php @workflow.signal async def approve plan self - None: self. plan approved = True @workflow.signal async def reject plan self, feedback: str = "" - None: self. plan approved = False self. plan feedback = feedback Signals only work while a Workflow is still running — once it's completed, failed, or been terminated, it can no longer receive one. Query reads a Workflow's current state, synchronously and without side effects: php @workflow.query def state self - WorkflowState: return WorkflowState status=self. status, project name=self. project name, preview url=self. preview url, Queries can also run against a completed Workflow, within your namespace's retention period. This doesn't extend to terminated Workflows, which aren't safely queryable. Because Temporal keeps a full event history, a Workflow can pause and resume without losing state — a pattern suited to human approval steps. python from temporalio import workflow @workflow.defn class ExampleWorkflow: @workflow.run async def run self : while True: self. plan approved = None await workflow.wait condition lambda: self. plan approved is not None if self. plan approved: break workflow.wait condition pauses execution until the given condition is true. Here, the Workflow waits until a Signal like approve plan above sets self. plan approved . If the process running this Workflow crashes while it's paused, a new Worker can pick it up and it resumes waiting at the same point. This example has no timeout, so it waits indefinitely. In production, you'd typically pass a timeout to wait condition and handle the case where no decision ever comes. This is part of an ongoing series on my channel https://youtube.com/@joxiahdev , Mastering Pydantic AI .