Introducing Kestrel Workflows

Every platform team has the same backlog: incident runbooks that live in someone's head, provisioning requests that pile up in Slack, CI/CD glue scripts that only one person knows how to run, and a long tail of "can you just" developer requests. Automating any of it usually means writing and maintaining brittle scripts, wiring up webhooks, and gluing together a dozen APIs by hand.

Today we're launching Kestrel Workflows — and the Workflow Agent that builds them. Describe what you want in plain English, and Kestrel assembles a complete, production-ready workflow: the trigger, the steps, the branching logic, and the approvals. AI builds the workflow once. From then on, it runs deterministically — the exact steps you defined, in the exact order, every single time.

From words to a workflow

Open the workflow builder and you'll see a single input at the bottom: Describe workflow in natural language. Click it to open the Workflow Agent and type what you want, for example:

Hit Generate and the agent produces a complete workflow graph — a name, a description, a configured trigger, and a directed graph of actions, conditions, and approval gates with the edges already wired together. It drops onto the canvas fully editable.

A canvas for everything you automate

Workflows are built as a visual DAG. A searchable sidebar lists every signal and action grouped by integration; drag blocks onto the canvas, connect them with edges, and click any node to configure its parameters. There are four kinds of blocks:

• Triggers

  The signal that starts a run — a Kubernetes pod crash, an AWS cost anomaly, a PagerDuty page, a PostHog exception, a Vercel deploy failure, a GitHub event, a custom webhook, or a developer request from Slack.

• Actions

  The work the workflow does — run RCA, generate and apply a fix, open a PR, create a Jira ticket, sync ArgoCD, run a Helm upgrade, post to Slack, and much more.

• Conditions

  An If / Else block that branches the workflow based on the value of a field from an earlier step.

• Approval gates

  Pause the workflow until someone signs off — in the platform, in Slack, or by merging a pull request.

Click to expand

Refine with words, drag-and-drop, or code

However you start, you can keep editing the same workflow three ways. Re-open the Workflow Agent and describe a change in natural language. Edit visually on the canvas — add or remove steps, draw new edges, tweak parameters and approvals. Or open the Code panel to see the equivalent Python SDK code for the workflow, edit it inline, copy it, or open a PR to manage it as code in your repo.

The visual canvas and the Python SDK are two views of the same workflow — switch between them however your team prefers to work.

Deterministic by design

The agent that builds your workflow is not the thing that runs it. Once a workflow is configured, execution is fully deterministic: it runs the exact steps you defined, in the exact order, with no model in the runtime loop deciding what to do next. No hallucinations, no surprises, no drift between runs.

That distinction is what makes Kestrel Workflows safe to put in front of production. AI gives you the speed of natural-language authoring; deterministic execution gives you the reliability your runbooks need.

Connected to your entire stack

Workflows ship with a large library of pre-built actions across your infrastructure, source control, CI/CD, observability, and communication tools — Kubernetes, AWS, OCI, GitHub, GitLab, ArgoCD, Helm, Slack, PagerDuty, Datadog, Jira, Linear, Confluence, Cloudflare, Vercel, PostHog, and more. For anything not yet built in, custom HTTP actions and custom webhook triggers let you wire up any service you run. You only need to connect each integration once — every workflow in your org shares the same connections.

That breadth is what lets a single workflow span an entire incident: detect a Kubernetes failure, run RCA, generate a fix, open a GitHub PR, wait for Slack approval, sync ArgoCD, and post the result back to the team — all in one deterministic run.

Built for more than incidents

Incident response is the obvious starting point, but Kestrel Workflows is a general-purpose automation engine for platform teams:

• Incident response — RCA, fix generation, restarts and scaling, PagerDuty alerts, postmortems.
• Cloud provisioning — generate and apply Kubernetes manifests, generate cloud resources, open IaC PRs, run Helm installs and upgrades.
• CI/CD pipelines — trigger GitHub Actions and GitLab pipelines, wait for runs, check status, sync and roll back ArgoCD.
• Developer requests — turn on-demand requests from Slack into safe, governed self-service workflows.
• Security — Cloudflare WAF rules, IP and ASN blocking, rate limits, and Zero Trust access policies, gated by approvals.

Ship it when you're ready

Every workflow starts as a draft. When you're ready, click Activate — Kestrel validates that it has a trigger, at least one action, and all required fields before it goes live. You can Pause at any time, set a cooldown to throttle how often a workflow re-fires, and enable failure alerts to a Slack channel or DM so you hear about problems immediately.

Getting Started

Open Workflows in your Kestrel dashboard and click New Workflow, then describe what you want to automate. Kestrel uses your connected integrations to wire the workflow up correctly — so the more you've connected, the more it can do out of the box. See the Create Workflows guide to learn every way to build — natural language, drag-and-drop, CLI, SDK, and MCP — and Integrations Setup to connect your stack.

This is the first in a series of releases for Kestrel Workflows. Over the coming weeks we'll cover access control and approvals, execution observability, the CLI, MCP server, and Python SDK, suggested workflows, the Workflow Assistant, and more.