Agentic Automation · Runs in your cluster

Codify the fix once. Run it safely, every time.

Turn the runbooks buried in your wiki into automation that actually runs. Build it without code, trigger it from chat, an alert or a schedule, and let it heal the routine problems on its own, behind an approval gate you control.

Book a Demo Start free

Self-hosted in your cluster · Human-approved by default · Crash-safe by design · Free forever tier

What is runbook automation?

Runbook automation turns a documented operational procedure into software that runs on demand or on its own. NudgeBee runbooks are built without code, triggered by chat, alert or schedule, and executed on a durable workflow engine so they never half-finish. Every change stays behind human approval unless you choose otherwise. When those runbooks run on their own, the result is auto remediation, sometimes called self-healing infrastructure: the routine problems get fixed without a person in the loop.

Automation that finishes the job
80+ built-in actionsKubernetes actions, cloud CLIs, database and queue calls, network diagnostics, scripting and control flow, ready to drop into a runbook.
4 ways to triggerBy hand, on a cron schedule, from an inbound webhook, or automatically off a NudgeBee event.
Crash-safeA runbook resumes exactly where it stopped and never runs a step twice, even if a node dies mid-execution.

Your runbooks are documents. They should be software.

  • The fix for the recurring 2am page is written down somewhere, in a wiki nobody fully trusts. Following it by hand is slow, and it drifts out of date the moment the environment changes.

  • The alternative is a drawer of shell scripts: undocumented, unversioned, and dangerous to run under pressure. One wrong flag on a tired night and the remediation becomes the incident.

  • What is missing is the middle ground: the procedure as real, versioned, testable software that runs the same way every time, with a human in the loop where it matters and out of it where it does not.

From trigger to fix, in three moves

  1. It triggers

    An alert, a schedule, a webhook or a NudgeBee event kicks the runbook off.

  2. It runs the steps

    Investigate with AI, run the commands across Kubernetes, AWS, Azure and GCP, and branch on what it finds.

  3. It closes the loop

    It pauses for a one-click approval on anything risky, then applies the fix, notifies Slack or email, opens a ticket or a pull request, and logs every step.

Worked example

When a pod is OOMKilled, a NudgeBee runbook investigates the cause, right-sizes the memory request, opens a pull request with the change, and posts the summary to Slack. It does nothing to production until someone clicks Approve.

The NudgeBee no-code runbook builder, wiring an OOMKilled alert through an AI investigation, an approval gate and a pull request on a drag-and-drop canvas.

Describe the fix. Nubi builds the runbook.

Tell the NudgeBee agent what you want in plain English, for example "when a pod is OOMKilled, right-size its memory and open a pull request after I approve," and it assembles the workflow for you.

Then refine it on a visual canvas, or keep it as YAML or JSON in git. The palette carries more than 80 built-in task types: Kubernetes actions (rightsize, restart, drain a node, delete a pod), the AWS, Azure and gcloud CLIs, database and message-queue queries, network diagnostics, and scripting in shell, Python, JavaScript or PowerShell.

Control-flow blocks (branch, loop, wait, approval, call another runbook) let a runbook make decisions, not just run a list.

  • 80+ built-in actions
  • Durable, crash-safe engine
  • Every run versioned
  • Exactly-once execution

Runbooks that finish what they start

Every runbook runs on a durable engine where each task is a checkpointed step. If a pod restarts or a node dies halfway through, the runbook resumes exactly where it left off and never executes a step twice.

Runbooks are versioned with a floating pointer to the live version, so you can change one safely while the current version keeps running.

This is the difference between a script that might leave your system half-migrated and a workflow that is crash-safe by construction.

A NudgeBee runbook run timeline showing completed steps, a step that resumed after a node restart, and the run paused at an approval gate before it changes anything.

Run it however the work arrives

Four triggers, so the runbook fires the way the problem actually reaches you.

By hand

From the console or a chat command, when a human decides.

On a schedule

With cron and a choice of overlap policies, so a slow run never stacks on top of itself.

From a webhook

Fanned out per integration, so an external system can kick off a fix.

Off a NudgeBee event

With filtering and lifecycle gating, so a specific finding triggers its own remediation.

Put the agent to work mid-runbook

A runbook step can call the NudgeBee agent itself. Mid-flow, a runbook can ask it to investigate a symptom, summarise the current state, or classify what it is looking at, and branch on the answer.

That is what makes these runbooks agentic rather than static: the automation can reason about what it finds and choose the next step, instead of blindly running to the end.

Investigate Summarise Classify Branch on the answer
A NudgeBee approval request in Slack asking a person to approve a memory right-size before the runbook applies it, with Approve and Reject buttons.

It reads freely. It changes nothing without you.

A runbook can pause and wait for a person. Approval steps are delivered where your team already works, and the runbook holds, durably, until someone approves or rejects, then continues or stops.

You decide which steps run on their own and which need a human hand on the switch, per runbook, per step.

Approvals arrive where the team already works:
  • Slack
  • Microsoft Teams
  • Google Chat
  • Email
  • Signed link

Let the routine ones heal themselves

This is auto remediation for the toil you already trust: the routine fixes become self-healing, running on a schedule inside the guardrails you set.

Auto-Pilot runs approved remediations on a schedule, autonomously, with real guardrails.

  • Blast-radius scored: before it applies a change, it checks the dependency graph and auto-applies only what is safe, and skips changes below a minimum-change threshold so it never churns tiny adjustments.
  • Dry-run first, so you see what it would do before it does it.
  • Resource filters, so the blast radius is exactly what you chose.
  • Skips itself if the in-cluster agent is disconnected, or if there is already an open pull request or ticket for that resource.
  • Change-gated notifications: it tells you when something actually changed, not on every quiet run.
  • Raises a GitOps pull request asynchronously, so the change lands through your normal review flow.

This is how the recurring, low-risk toil (rightsizing, cleanup) gets handled without a human, while everything else stays gated.

A NudgeBee Auto-Pilot dry-run listing proposed right-sizing changes for three workloads, with guardrail toggles and a note that it would raise three pull requests on the next live run.

It drives the tools you already run

Runbook tasks call these systems directly, and approvals and notifications flow back to the channels your team already lives in.

A runbook step can call any of these directly, and the approval that gates it comes back through the same channel your team is already watching.

Browse all integrations

Runs in your cluster. Nothing leaves.

Self-hosted

Runs inside your own Kubernetes cluster. Zero telemetry.

Outbound only

The in-cluster agent dials out over one WebSocket. No inbound ports, no exposed API server, no VPN.

Encrypted and signed

Credentials encrypted at rest. Agent messages signed.

Bring your own model

The AI-in-workflow steps run on the model provider you choose.

Nothing leaks to the model

Outbound AI requests are scanned for secrets before they reach any model, and can be redacted or blocked.

Source-available

Source-available and self-hosted, with a free Community edition.

See it under real incident load

Multicloud complexity does not slow down when your team does. NudgeBee's SRE and AIOps agents absorb routine operations autonomously so our teams focus on high-value engineering. Their extensive library of specialist agents gives Rackspace the building blocks to rapidly create custom agentic workflows at scale.
Nirmal Ranganathan CTO, Global Public Cloud, Rackspace
A busy retail checkout counter, standing in for a high-traffic e-commerce platform
Runbook automation case study

Top-10 e-commerce platform ends $180K/hr peak-season outages with agentic AI-SRE

3,000+ daily alerts correlated into actionable incidents, automated runbooks executed, and 61% of issues resolved without a human ever getting paged.

Read the case study

How this compares

Measured against the category of script and job runners teams usually reach for first, not against any one product.

NudgeBee runbook automation compared with the general category of script and job runners, across seven capabilities.
Capability Script / job runners NudgeBee
No-code visual builder Sometimes Yes
Crash-safe, exactly-once execution Rarely Yes
Trigger from chat, alert, schedule and event Partial Yes, all four
AI steps inside the workflow No Yes
Human approval gates built in Add-on Yes, native
Autonomous auto-pilot with dry-run and guardrails No Yes
Self-hosted in your own cluster Varies Yes

Compared against the category of runbook and workflow automation tooling. Individual products vary.

Runbook automation FAQ

Frequently asked questions

For the engineer who will build the runbook and the lead who has to sign off on what it can touch.

Book a Demo
Runbook automation turns a documented operational procedure into software that runs on demand or on its own. Instead of a person following steps in a wiki, the procedure becomes a versioned workflow that can be triggered, executed and audited, with every risky step held behind a human approval you control.
A script runs top to bottom and is lost if the machine running it dies. A NudgeBee runbook is built without code, versioned on every change, resumable after a crash, able to call the AI agent mid-flow, and able to pause for a human approval before it touches anything.
No. Describe what you want to the NudgeBee agent in plain English and it assembles the workflow for you. From there you refine it on a visual canvas using a palette of ready-made tasks, or keep the same runbook as YAML or JSON in git if you would rather version it that way.
More than 80 built-in task types. Kubernetes actions such as rightsize, restart, drain a node and delete a pod, the AWS, Azure and gcloud CLIs, database and message-queue queries, network diagnostics, scripting in shell, Python, JavaScript or PowerShell, and control flow: branch, loop, wait, approval and calling another runbook.
Four triggers. By hand from the console or a chat command, on a cron schedule with a choice of overlap policies, from an inbound webhook fanned out per integration, or automatically off a NudgeBee event with filtering, so a specific finding kicks off its own remediation.
Only where you configure Auto-Pilot. Auto-Pilot runs approved remediations on a schedule with dry-run, resource filters, and guardrails that skip execution if the in-cluster agent is disconnected or a pull request is already open for that resource. Everything outside Auto-Pilot waits for explicit human approval.
It resumes. Every task is a checkpointed step on a durable, crash-safe engine, so if a pod restarts or a node dies mid-execution the runbook picks up exactly where it stopped. No step is executed twice, and nothing is left half-applied.
Yes. A step can call the NudgeBee agent mid-flow to investigate a symptom, summarise the current state, or classify what it is looking at, and the runbook branches on the answer. That is what makes these runbooks agentic rather than a fixed list of commands.
Yes. NudgeBee is self-hosted inside your own Kubernetes cluster with zero telemetry. The in-cluster agent dials out over a single WebSocket, so there are no inbound ports and no exposed API server. Credentials are encrypted at rest and agent messages are signed.
There is a free forever tier covering up to two clusters or cloud accounts, with no credit card required. It is source-available and self-hosted, so you can run it in production internally on the free Community edition. Paid plans and the full breakdown are on the pricing page.
Yes. When you let a runbook run on its own through Auto-Pilot, NudgeBee performs automated remediation: it applies the routine fix without a human, on a schedule, with dry-run and guardrails. Everything outside that stays behind approval, so self-healing is scoped to exactly what you allow.

They find the problem and propose the fix. This is the layer that runs it, on one platform and one knowledge graph.

AI SRE Investigates incidents to a cited root cause, with the evidence attached. Explore AI FinOps Finds cloud and Kubernetes waste and attaches the exact fix. Explore Kubernetes Operations Day-two clusters: live topology, alert de-noising and safe upgrades. Explore CloudOps One AI console across AWS, Azure and GCP. Explore
Free forever · No credit card

Turn your next 2am page into a runbook.

Free forever on up to 2 clusters. No credit card. Self-hosted, so nothing leaves your environment.

Start free Book a demo

Read the source on GitHub