Skip to main content
For the complete documentation index for agents and LLMs, see llms.txt.

Deploy a Pipeline

After you create and save a pipeline, deploy it to test it in the Playground. Deploying a pipeline is useful for internal testing and debugging where endpoint stability is not a priority. Deploy always uses the latest version of the pipeline, so you can deploy without choosing between drafts and versions.

About This Task​

You can run the pipeline in Builder without deploying it. This is useful for quick testing and debugging. To share a pipeline or try it out in the Playground, you must deploy it first.

About This Task​

When you deploy a new version of a pipeline or redeploy an existing one, the version that's already running keeps handling queries until the new version deploys successfully. Your pipeline stays available the whole time. If the new deployment fails, the current version keeps running, so there's no downtime.

Deploy a Pipeline from the UI​

Check Issues Before Deploying

Before deploying, open your pipeline in Builder and check the Issues tab at the bottom of the canvas. It shows all validation problems that could cause the deployment to fail. Click Inspect next to any issue to fix it first, or click Fix with AI to open the AI assistant, which reasons through the fix and can apply it for you.

  1. In Haystack Enterprise Platform, click Pipelines.
  2. Go to Drafts, find the pipeline you want to deploy, and click its name.
  3. In Builder, click Publish.
  4. Check the version you're deploying and click Continue.
  5. Choose to deploy the pipeline without a service and click Continue.

Deploy a Pipeline with the REST API​

Here's the code you can use. For more information about this endpoint, see Deploy Pipeline API.

You need to Generate an API Key first.

Copy this code replacing workspace name, pipeline name, and API key:

curl --request POST \
     --url https://api.cloud.deepset.ai/api/v1/workspaces/<workspace_name>/pipelines/<pipeline_name>/deploy \
     --header 'Accept: application/json' \
     --header 'Authorization: Bearer <your_API_key>'

What To Do Next​

It may happen that the deployment fails, and the pipeline status is failed to deploy. This status means the pipeline wasn't deployed. If you encounter this error, you must take some troubleshooting action.

To view pipeline information, such as statistics or feedback, click the pipeline name, and go to Analytics.

Deployed pipelines that haven't been used for a while become idle to save resources. The default idle timeout depends on the pipeline service level. By default, production pipelines go on standby after 24 hours of not being used and development pipelines after 20 minutes. You can customize the idle timeout in pipeline settings. An idle pipeline gets automatically activated if you run a search with it or you can trigger the activation manually on the Pipelines page. For more information, see Pipeline Hosting Settings.