Shop is live

We build Workstation AI machines for websites, AI agents, and LLM training — configure and buy online.

Get Trade Prices
AI Products
AI LabsOpenAI AgentsClaude AgentsGrok BotWorkstation AI native CRMMarketingAll Products
AI Solutions
Applied AIAI AgentsGenerative AIAI AutomationAI UpskillingAI SME PackagesEnterprise AI LabAI WorkstationsNVIDIA Hardware & SoftwareNVIDIA Accelerated ComputingPrivate AIGPU Private CloudGPU ClustersEdge AIAI by Industry
AI Services
Platform ModernisationDigital EngineeringData Foundations & AIAutonomous OperationsAI ConsultancyDevOps AutomationCyber SecuritySoftware DevelopmentAgent BuildingMLOps SetupGPU LeasingGPU Private Cloud
About Us
CompanyOur StoryPartnersCustomer Stories
AI Data Centres
AI Data CentresGPU Private CloudGPU ClustersGPU Leasing
AI InsightsJev seriesBenchmark LabModel selection
Technical Notes
Documentation
Linux OSUbuntuRed Hat Enterprise LinuxSUSE LinuxAlpine LinuxDebian, Rocky & AlmaLinux
AI Cloud PlatformsAWS & Amazon BedrockGoogle Cloud & Vertex AIMicrosoft Azure & AI FoundryGPU clouds & marketplaces
Private AI Data CentresFor law firmsFor accounting firmsOn-premises platformsFully secured cloud
ToolsRing PromoterJobshout
News
Blog
Contact UsSign in

AI workstations, AI Multi Agentic Software, GPU infrastructure, and intelligent agent solutions for modern businesses.

Contact Us

AI Solutions

Applied AIAI AgentsGenerative AIAI AutomationAI UpskillingAI SME PackagesEnterprise AI LabAI by Industry

Products

All ProductsWorkstation AI native CRM & ERPMarketingOpenAI AgentsRing PromoterJobshout

Company

About UsCompanyAI Data CentresWhy WorkstationPartnersCustomer StoriesPricingContactTenthMatrix

Resources

AI InsightsTechnical NotesDocumentationLinux OSAI Cloud PlatformsPrivate AI Data CentresBlogSearchSitemap
UK Head Office
77-79 Marlowes, Hemel Hempstead HP1 1LFDirections - Take Junction 20 off M25 Outer LondonCompany No: 11641870Mon - Fri: 9:00 AM - 6:00 PM GMT
+44 7515 356 146
Belgium Office
Workstation SRL, Rue Vanderkindere 34, 1180 Uccle, BrusselsBE 0751.518.683Mon - Fri: 9:00 AM - 6:00 PM CET
+32 492 45 67 46
India Office
#159 Sector 9, Pocket 1, DDA Flats, 110077 Dwarka, New Delhi
+91 98881 98841
Taiwan Office
3F, No. 2-4, Aisi Rd., Ren'ai Dist., Keelung City 200, Taiwan基隆市仁愛區愛四路2之4號3樓Mon - Fri: 9:00 AM - 6:00 PM Taiwan time (GMT+8)
+886 926 892 701+32 492 45 67 46
Estonia Office
Aleksander Puškini tn 30-25, 20306 Narva, EstoniaMon - Fri: 9:00 AM - 6:00 PM Estonia time (EET)
+32 492 45 67 46

© 2026 Workstation AI. All rights reserved.

PrivacyData DeletionCookiesTerms of ServiceWebsite Sitemap

NVIDIA, the NVIDIA logo, RTX, RTX PRO, Blackwell, CUDA, NVLink, MGX, Omniverse, Spectrum-X, Quantum, BlueField, DGX and HGX are trademarks and/or registered trademarks of NVIDIA Corporation in the US and other countries. AMD, the AMD Arrow logo, EPYC, Instinct, Ryzen and Threadripper are trademarks of Advanced Micro Devices, Inc. Dell and PowerEdge are trademarks of Dell Inc. or its subsidiaries. Intel and Xeon are trademarks of Intel Corporation or its subsidiaries. AMAX, AceleMax and LiquidMax are trademarks of AMAX Engineering Corporation. All other product names, logos and brands are the property of their respective owners. We use them only to identify the products we describe, supply or support; their use does not mean their owners endorse us.

Home / Technical Notes / Technology
KubernetesDevOpsAIOpsOpen Source

KubePilot: CoPilot, Pilot & AutoPilot for Faster Kubernetes Incidents

Technical brief: three-mode incident loop, install (source/Helm/Docker/iOS), AutoPilot safety rails, MCP, runbooks, and production checklist

August 12, 2026Technology5 min read

KubePilot is an open-source Kubernetes troubleshooting cockpit built by the Workstation engineering team. It combines AI-assisted diagnosis (CoPilot), SRE-style cluster browsing (Pilot), and policy-gated remediation (AutoPilot) so platform, SRE, and DevOps teams move from noisy signals to a safe fix — without juggling kubectl tabs mid-incident. Product site: kubepilot.org. Source: github.com/bwalia/kubepilot.

KubePilot cover — CoPilot Pilot AutoPilot

Navigation. Companion blog: business summary. Related: GitOps with Argo CD & Flux, Kubeflow + Argo CD, LLMs on Kubernetes, MCP OAuth & Vault.
Agent digest.
  • What: open-source K8s cockpit — CoPilot (AI RCA) + Pilot (SRE browser) + AutoPilot (gated fixer).
  • Who: SRE, DevOps, platform / AIOps teams who need faster MTTR with auditable automation.
  • Run: binary serve on :8383, Helm chart, Docker (ghcr.io/kubepilot/kubepilot), iOS companion.
  • AI: local LLMs via Ollama; Copy Prompt; MCP server on :9090.
  • Safety: read-only Pilot defaults; AutoPilot off/dry-run/active; confidence floors; blocklists; kill switch; decision ledger.

1. Watch the walkthrough

youtu.be/CsiJWpfncqA — Balinder Walia, Workstation.

2. The problem KubePilot solves

Modern clusters emit more signal than humans can triage in a single page: CrashLoops, ImagePullBackOff, OOMKills, node pressure, flaky probes, and YAML drift. Typical response spreads across:

  • kubectl history and shell scripts,
  • a browser UI (Lens / Rancher / dashboard),
  • logs in a third tool,
  • an LLM chat with pasted YAML (and no cluster auth).

That fragmentation burns minutes when minutes are the SLA. KubePilot collapses the loop into one cockpit: spot → ask → verify → fix.

Three-mode incident loop

3. Mode A — CoPilot (AI-assisted troubleshooting)

CoPilot is the natural-language and automated RCA surface:

  • Ask operational questions in plain English (“why is checkout CrashLooping in prod?”).
  • One-click AI Analyze on problem pods and events.
  • Root-cause analysis with evidence chains, confidence, risk, suggested commands, YAML fixes, and next steps.
  • Anomaly detection for CrashLoop, OOM, ImagePull, and node pressure patterns.
  • Copy Prompt to reuse the exact diagnosis prompt with your preferred model.
  • Designed for Ollama / local models so cluster context can stay on your network.

Cluster events and AI troubleshooting

4. Mode B — Pilot (SRE troubleshooting browser)

Pilot is the Lens/Rancher-style read-only workbench for verification before mutation:

  • Browse Deployments, StatefulSets, DaemonSets, Jobs, Services, Ingresses, ConfigMaps, Secrets, PVCs.
  • Pod workbench: overview, containers, events, logs, sanitized YAML, restart counts, uptime.
  • Live cluster resource gauges (CPU, memory, disk) — Longhorn-aware where applicable.
  • Service topology canvas: Ingress → Service → Workload → Pod with status colours and ports.
  • Node LAN, WAN, and tunnel (e.g. WireGuard / flannel) IP classification — not only the overlay address.
  • Port-forward tunnels from the UI; multi-cluster kubeconfig upload and context switching without restart.

Pilot dashboard overview with gauges and node IPs

5. Mode C — AutoPilot (AI fixer with safety rails)

AutoPilot closes the loop with auditable self-healing — never “YOLO apply” as the default:

Control Purpose
Modes: off / dry-run / activePreview before apply; keep automation parked when needed
Confidence floorsSkip low-confidence RCAs instead of guessing
Namespace blocklists & action allow-listsProtect kube-system / prod namespaces; limit mutation types
Cooldowns & hourly capsPrevent remediation storms
Kill switchOne-click pause for operators
Decision ledgerExecuted, skipped, escalated, failed — audit trail
CR-code escalationHuman approval for risky actions

AutoPilot anomaly to audited remediation workflow

6. Full feature map (from kubepilot.org)

  • Pre-built runbooks: seven opinionated diagnostic workflows; custom YAML runbooks with fsnotify hot-reload (user IDs override builtins).
  • Durable RCA history: optional embedded SQLite (WAL, no CGO) with retention for reports and anomalies across restarts.
  • Alerts & timeline: warnings, events, anomalies, RCA as a searchable ops timeline; optional Slack incident cards.
  • AI health chip: know Ollama is ready before an incident depends on it.
  • Security-first defaults: optional auth, read-only browsing, mutation gates, CORS policies.
  • MCP agent protocol: built-in MCP server for multi-cluster agent orchestration and programmatic access.
  • Native iOS companion: SwiftUI app — health cards, pods, logs, RCA, Face ID, widgets, Siri / App Intents.

7. Install options

Prerequisites for source builds: Go 1.22+, Node.js 18+, a reachable cluster + kubeconfig; Ollama recommended for local AI.

7.1 Build from source

git clone https://github.com/bwalia/kubepilot.git
cd kubepilot
make dashboard-install
make dashboard
make build
KUBEPILOT_KUBECONFIG="$HOME/.kube/config" ./dist/kubepilot serve --dashboard-port=8383
# Dashboard/API: http://localhost:8383
# MCP server: :9090

7.2 Helm (cluster install)

helm repo add kubepilot https://bwalia.github.io/kubepilot
helm repo update
helm upgrade --install kubepilot kubepilot/kubepilot \
  -n kubepilot --create-namespace
kubectl port-forward svc/kubepilot -n kubepilot 8080:8080
# Open http://localhost:8080

Also supported: NodePort, Ingress (+ cert-manager TLS), LoadBalancer, and values overrides — see the chart README on GitHub.

7.3 Docker

docker run --rm -p 8383:8383 -p 9090:9090 \
  -v "$HOME/.kube:/root/.kube:ro" \
  ghcr.io/kubepilot/kubepilot:latest \
  serve --dashboard-port=8383

7.4 iOS companion

cd ios
brew install xcodegen
./generate.sh
open KubePilot.xcodeproj
# Point the app at http://<server-ip>:8383

8. Recommended operating model

  1. Day 0: run Pilot read-only against a non-prod cluster; wire Ollama; confirm AI health chip.
  2. Day 1: use CoPilot AI Analyze on known bad pods; compare RCA to your runbooks; tune Copy Prompt into your team LLM.
  3. Day 2: enable AutoPilot in dry-run only; review the decision ledger for a week.
  4. Day 3+: promote selected namespaces to active with tight allow-lists, confidence floors, and Slack alerts; keep kube-system blocked.
  5. Always: kill switch rehearsed; CR-code path for risky mutations; MCP tools behind auth (see our agentic security brief).

9. Production checklist

  • ☐ Auth enabled for any network-reachable dashboard.
  • ☐ AutoPilot starts in dry-run; active only after ledger review.
  • ☐ Namespace blocklist includes system / shared namespaces.
  • ☐ Hourly caps and cooldowns set below storm thresholds.
  • ☐ Ollama (or approved LLM path) monitored via AI health.
  • ☐ Slack or pager integration for high-severity anomalies.
  • ☐ GitOps still owns desired state — AutoPilot is a tactical fixer, not a substitute for Argo CD / Flux.
  • ☐ Star and track releases: github.com/bwalia/kubepilot.

10. Why Workstation ships this open source

Workstation builds automation platforms for real ops teams. KubePilot is our bet that AIOps only works when diagnosis and remediation share one evidence trail — and when automation is gated like a human SRE would gate a change. We publish it under open source so platform teams can run it on their own networks, extend runbooks, and wire MCP agents without vendor lock-in.

Start at kubepilot.org, clone the repo, and watch the walkthrough on YouTube. Companion summary: blog post.

Share this article

More in Technology

AI Developer Laptops: Running LLMs Locally on Linux with 16 GB and 24 GB GPUs (2026 Guide)

AI Developer Laptops: Running LLMs Locally on Linux with 16 GB and 24 GB GPUs (2026 Guide)

What fits on an RTX 5080 or RTX 5090 Laptop GPU, the memory arithmetic, the open models worth running, Ubuntu 26.04 setup with the NVIDIA stack, Ollama, llama.cpp and vLLM, coding assistants, LoRA fine-tuning, security, and when to move work to a server

Read more
Testing ML and LLM Systems You Can't Prove Correct: A Python Evidence Kit

Testing ML and LLM Systems You Can't Prove Correct: A Python Evidence Kit

Evidence instead of proofs, in about 400 lines of Python with real output: slice metrics with bootstrap intervals and a pass/fail contract, Hypothesis property and metamorphic tests, calibration and conformal prediction, Pydantic contracts and regression evals for LLM output, drift checks, a CI gate, and exact verification of a small neural network

Read more
TorchCurves in Practice: Learnable B-spline and Legendre Curves in PyTorch

TorchCurves in Practice: Learnable B-spline and Legendre Curves in PyTorch

Install torchcurves, fit and size a curve, choose B-spline or Legendre and the input map, build a monotone calibration curve and an additive model on tabular data, compare with SciPy, scikit-learn, a KAN and an MLP, export coefficients to JSON, test and run it in production, with real outputs

Read more