The problem

A pod is stuck in CrashLoopBackOff. So you run kubectl describe pod …, then kubectl logs … --previous, then kubectl get events …, then squint at the exit code and piece together what went wrong. Every time.

kubectl-ai-explain does that legwork for you:

$ kubectl ai-explain checkout-api-6b9c7d -n prod

❌ Summary: The container is crashing on startup because it can't reach the database.
🔎 Likely cause (high confidence): DB_HOST is unset, so the app exits with code 1 before serving traffic.

📋 Evidence:
  • Container state is CrashLoopBackOff, exit code 1, 7 restarts
  • Logs: "SQLSTATE[HY000] [2002] Connection refused"

🔧 Suggested fixes:
  $ kubectl set env deployment/checkout-api DB_HOST=mysql-service
  $ kubectl rollout status deployment/checkout-api

✨ Features

• 🩺 One command - gathers the pod's status, events, and recent logs, and explains the failure.
• 🔌 Provider-agnostic - Claude (default), OpenAI, or a local model via Ollama (no API key, fully private).
• 🔒 Secret-aware - best-effort redaction of passwords/tokens/keys before anything is sent to an LLM, plus a --dry-run to see exactly what would be sent.
• 🧩 Native kubectl plugin - invoke as kubectl ai-explain <pod>.
• 📤 Scriptable - --json for machine-readable output.
• 🚫 Read-only - it suggests fix commands; it never runs them.

📦 Installation

From source (works today):

git clone https://github.com/jay-tank/kubectl-ai-explain.git
cd kubectl-ai-explain
pipx install .            # recommended — lands on PATH as a kubectl plugin
#   …or:  pip install --user .

Install a provider SDK for whichever backend you'll use (Ollama and the built-in mock need none):

pip install --user '.[claude]'    # Anthropic Claude
pip install --user '.[openai]'    # OpenAI

📦 A published PyPI release (pipx install kubectl-ai-explain) is planned — until then, install from source as above.

Because the executable is named kubectl-ai-explain, kubectl auto-discovers it:

kubectl ai-explain --help     # works as a plugin
kubectl plugin list           # should list kubectl-ai-explain

⚡ Quick start

# Default provider is Claude — set your key (see PROVIDERS.md), then:
export ANTHROPIC_API_KEY=...          # your key, your environment
kubectl ai-explain my-pod -n prod

# No API key? Run a local model with Ollama:
kubectl ai-explain my-pod --provider ollama

# Just want to see what would be sent to the LLM (no call made)?
kubectl ai-explain my-pod --dry-run

🔧 Usage

kubectl ai-explain <pod> [options]

  -n, --namespace     Namespace (default: default)
      --container     Limit to a single container
      --context       kubectl context to use
      --tail          Log lines per container (default: 50)
      --provider      claude | openai | ollama | mock   (default: claude)
      --model         Model override (or $AI_EXPLAIN_MODEL)
      --json          Machine-readable JSON output
      --no-color      Disable colored output
      --dry-run       Print the redacted prompt and exit (no LLM call)

Full guide: docs/USAGE.md · provider setup: docs/PROVIDERS.md.

🔒 Privacy

To explain a failure, the tool sends the pod's status, events, and recent log tail to your chosen LLM. Before sending, it redacts common secret patterns (passwords, tokens, AWS keys, bearer tokens). Two extra guards:

• --dry-run prints the exact, redacted payload so you can review it first.
• --provider ollama keeps everything on your machine — nothing leaves your network.

Redaction is best-effort, not a guarantee — for sensitive clusters, prefer Ollama. See docs/PROVIDERS.md.

🧠 How it works

kubectl ai-explain <pod>
   └─ collector   → kubectl get pod -o json | get events | logs --previous
   └─ prompt      → distilled, secret-redacted, token-bounded context
   └─ provider    → claude | openai | ollama | mock   (pluggable)
   └─ render      → terminal report  (or --json)

🤝 Contributing

Contributions welcome — see CONTRIBUTING.md. Run the tests with:

pip install -e '.[dev]' && pytest

📄 License

MIT — see LICENSE.