The Workflow Assistant

Most teams already have their operational knowledge written down somewhere — a runbook in Confluence, a Notion page, a shell script in a repo that someone runs by hand during incidents. Suggested Workflows turns your integrations into automations. The Workflow Assistant turns your knowledge into automations.

It's a conversational agent that helps you build workflows from a description, a question, or your existing runbooks and scripts — and hands you back a production-ready workflow you can review, refine, and ship.

Click to expand

Just ask

Open the assistant and describe what you want to automate, or ask it what's possible. Things like:

• "What can I automate for cloud provisioning and developer requests?"
• "Help me automate our on-call escalation process."
• "What can I automate for Kubernetes pod failures?"

The assistant reasons about your connected integrations and your description, then proposes a complete workflow — with a name, a description, and a visual preview of the workflow graph right in the conversation.

If your description is missing details a valid Kestrel workflow needs — which cluster or account to target, a Slack channel to post to, a Jira project, an approval threshold — the assistant asks follow-up questions instead of guessing. It only finalizes once it has everything required, so what it hands back is always a complete, valid workflow you can deploy.

Bring your runbooks and scripts

This is where it gets powerful. Upload a runbook or script and ask the assistant to generate workflows from your existing procedures — or point it at scripts in your connected GitHub or GitLab repos that you're still running manually. You can also pull in context from Confluence, Notion, and Glean so the workflow reflects how your team actually operates.

Upload an ArgoCD deployment rollback runbook, for example, and the assistant produces a complete workflow that detects the failed deployment, investigates it, rolls it back, and notifies the right people — translating the steps a human used to follow into a deterministic automation.

Save it, refine it, or ship it as code

Once the assistant generates a workflow, you choose what happens next, right from the preview:

• Save to Library

  Keep it for later alongside your other suggested and created workflows.

• Create as Draft

  Open it on the workflow builder canvas to refine the steps, parameters, and approvals before activating.

• Copy the Python SDK code

  Grab the equivalent Python SDK code if you'd rather deploy it programmatically or check it into a repo.

Getting Started

Open the Workflow Assistant from the Workflow Library, then describe a process or upload a runbook. The more you've connected — integrations, repositories, and knowledge sources — the more accurately it can turn your existing procedures into workflows.