Nexus Python Quickstart

Temporal Nexus connects Temporal Applications within and across Namespaces using a Nexus Endpoint, a Nexus Service contract, and Nexus Operations. Build a Nexus Service that wraps an existing Temporal Workflow, then invoke it from a caller Workflow.

NEW TO NEXUS?

This page will help you get a working sample running in Python. To evaluate whether Nexus fits your use case, see the evaluation guide and to learn more about Nexus features, click here.

Prerequisites: Complete the Python SDK Quickstart first. You should have activities.py, workflows.py, worker.py, and starter.py from that guide.

What you'll build

You have SayHelloWorkflow running in the default Namespace. By the end of this guide:

1. A Nexus Service will expose SayHelloWorkflow as an Operation.
2. A second Namespace will contain a Workflow that calls that Operation.
3. The caller Workflow will get back "Hello Temporal" — the same result, but across Namespaces.

1. Define the Nexus Service

Create a file called service.py that defines the Nexus Service contract.

Creating a Nexus Service establishes the contract between your implementation and any callers. It provides type safety when invoking Nexus Operations and ensures that operation handlers fulfill the contract.

The @nexusrpc.service decorator declares a service with typed operations. SayHelloWorkflow returns str, so the operation output type is str.

USE AN IDL TO GENERATE SERVICE DEFINITIONS

You can also define your Nexus Service contract using an IDL and generate the Python service definitions with nexus-rpc-gen.

from dataclasses import dataclass

import nexusrpc


@dataclass
class MyInput:
  name: str


@nexusrpc.service
class SayHelloNexusService:
  say_hello: nexusrpc.Operation[MyInput, str]

2. Define the Nexus Operation handlers

Create a file called handler.py that implements the Nexus Operation handler.

Operation handlers contain the logic that runs when a caller invokes a Nexus Operation.

The @nexus.workflow_run_operation decorator creates an asynchronous Nexus Operation that starts a Workflow. The handler bridges the Nexus MyInput dataclass to SayHelloWorkflow's str parameter by extracting input.name.

import uuid

import nexusrpc.handler
from temporalio import nexus

from service import MyInput, SayHelloNexusService
from workflows import SayHelloWorkflow


@nexusrpc.handler.service_handler(service=SayHelloNexusService)
class SayHelloNexusServiceHandler:
  @nexus.workflow_run_operation
  async def say_hello(
      self, ctx: nexus.WorkflowRunOperationContext, input: MyInput
  ) -> nexus.WorkflowHandle[str]:
      return await ctx.start_workflow(
          SayHelloWorkflow.run,
          input.name,
          id=f"say-hello-nexus-{uuid.uuid4()}",

          # Task queue defaults to the task queue this Operation is handled on.
      )

3. Register the Nexus Service handler in a Worker

Update your existing worker.py to register the Nexus Service handler.

A Worker will only poll for and process incoming Nexus requests if the Nexus Service handlers are registered. This is the same Worker concept used for Workflows and Activities.

The nexus_service_handlers parameter registers the handler so it can receive Nexus Operation requests.

import asyncio

from temporalio.client import Client
from temporalio.worker import Worker

from workflows import SayHelloWorkflow
from activities import greet
from handler import SayHelloNexusServiceHandler


async def main():
  client = await Client.connect("localhost:7233")
  worker = Worker(
      client,
      task_queue="my-task-queue",
      workflows=[SayHelloWorkflow],
      activities=[greet],
      nexus_service_handlers=[SayHelloNexusServiceHandler()],
  )
  print("Worker started.")
  await worker.run()


if __name__ == "__main__":
  asyncio.run(main())

4. Develop the caller Workflow

Create a file called caller.py that defines a Workflow which invokes the Nexus Operation.

The caller Workflow demonstrates the consumer side of Nexus. Instead of importing handler code directly, the caller only depends on the Service contract. This keeps the caller and handler decoupled so they can live in separate Namespaces, repositories, or even teams.

The workflow.create_nexus_client() method creates a client bound to your Nexus Service and Endpoint. execute_operation starts the operation and waits for the result.

from datetime import timedelta
from temporalio import workflow
from service import MyInput, SayHelloNexusService

NEXUS_ENDPOINT = "my-nexus-endpoint-name"


@workflow.defn
class CallerWorkflow:
  @workflow.run
  async def run(self, name: str) -> str:
      nexus_client = workflow.create_nexus_client(
          service=SayHelloNexusService,
          endpoint=NEXUS_ENDPOINT,
      )
      return await nexus_client.execute_operation(
          SayHelloNexusService.say_hello,
          MyInput(name=name),
          schedule_to_close_timeout=timedelta(seconds=10),
      )

5. Create the caller Namespace and Nexus Endpoint

Before running the application, create a caller Namespace and a Nexus Endpoint to route requests from the caller to the handler. The handler uses the default Namespace that was created when you started the dev server.

Namespaces provide isolation between the caller and handler sides. The Nexus Endpoint acts as a routing layer that connects the caller Namespace to the handler's target Namespace and Task Queue. The endpoint name must match the variable defined in caller.py from step 4.

Make sure your local Temporal dev server is running (temporal server start-dev).

temporal operator namespace create --namespace my-caller-namespace

temporal operator nexus endpoint create \
--name my-nexus-endpoint-name \
--target-namespace default \
--target-task-queue my-task-queue

6. Run and Verify

Create a file called caller_starter.py to start the caller Worker and execute the Workflow.

This step brings everything together: the caller Worker hosts CallerWorkflow, which uses the Nexus client to invoke say_hello on the handler side. The full request flows from the caller Workflow, through the Nexus Endpoint, to the handler Worker running SayHelloWorkflow, and back to the caller.

Run the application:

1. Start the handler Worker in one terminal:

source env/bin/activate
python3 worker.py

2. Run the caller in another terminal:

source env/bin/activate
python3 caller_starter.py

You should see:

Workflow result: Hello Temporal

Open the Temporal Web UI and find the CallerWorkflow execution. You should see NexusOperationScheduled, NexusOperationStarted, and NexusOperationCompleted events in the Workflow history.

import asyncio
import uuid

from temporalio.client import Client
from temporalio.worker import Worker

from caller import CallerWorkflow

CALLER_TASK_QUEUE = "my-caller-task-queue"
NAMESPACE = "my-caller-namespace"


async def main():
  client = await Client.connect("localhost:7233", namespace=NAMESPACE)

  async with Worker(
      client,
      task_queue=CALLER_TASK_QUEUE,
      workflows=[CallerWorkflow],
  ):
      result = await client.execute_workflow(
          CallerWorkflow.run,
          "Temporal",
          id=f"caller-workflow-{uuid.uuid4()}",
          task_queue=CALLER_TASK_QUEUE,
      )
      print("Workflow result:", result)


if __name__ == "__main__":
  asyncio.run(main())

Next Steps

Now that you have a working Nexus Service, here are some resources to deepen your understanding:

• Python Nexus Feature Guide: Covers synchronous and asynchronous Operations, error handling, cancellation, and cross-Namespace calls.
• Nexus Operations: The full Operation lifecycle, including retries, timeouts, and execution semantics.
• Nexus Services: Designing Service contracts and registering multiple Services per Worker.
• Nexus Patterns: Comparing the collocated and router-queue deployment patterns.
• Error Handling in Nexus: Handling retryable and non-retryable errors across caller and handler boundaries.
• Execution Debugging: Bi-directional linking and OpenTelemetry tracing for debugging Nexus calls.
• Nexus Endpoints: Managing Endpoints and understanding how they route requests.
• Temporal Nexus on Temporal Cloud: Deploying Nexus in a production Temporal Cloud environment with built-in access controls and multi-region connectivity.