Local-first
kind + Ollama. Your laptop is the production environment we operate — and break. Cloud is opt-in.
srebudo.ai
An old sensei. A broken cluster. Ten chapters. You leave with an agent CLI you built yourself — and the scars to prove it works.
A young engineer asks: “Sensei, which framework should I learn?”
Budo replies: “First, write the loop. The framework is what remains when you understand why you no longer want to write the loop.”
What this is
A hands-on course for experienced SREs. Every chapter you build one real agent against a real running system — a kind cluster running a 12-service shop with real metrics, real logs, and failures injected by us. Every agent becomes a subcommand of one evolving CLI: budo. By the black belt, budo is an interactive terminal agent in the spirit of opencode and Claude Code — running on a local model.
From scratch, then SDK
Ch1–4: raw loop, no frameworks. Ch5+: migrate to the Claude Agent SDK and feel exactly what it buys you.
Real failure classes
Silent env-var typos, flaky tests, noisy neighbors, IAM drift, prompt injection in your own logs, runaway bills.
Break it, harden it
Every chapter ends with an attack on the agent you just built. The fix is the lesson.
The belts
| Belt | Chapter | The agent | budo gains |
|---|---|---|---|
| — | Ch0 The Dojo | the lab itself | just lab-up |
| ⬜ | Ch1 The Naked Loop | log triage, zero frameworks | budo logs |
| 🟨 | Ch2 IaC | Terraform plan reviewer | budo plan-review |
| 🟧 | Ch3 CI/CD | pipeline failure root-causer | budo ci |
| 🟩 | Ch4 Observability | the PromQL whisperer | budo why |
| 🟦 | Ch5 Oncall | incident copilot | budo oncall + REPL |
| 🟪 | Ch6 Scale | capacity planner that opens PRs | budo capacity |
| 🟫 | Ch7 Cost | FinOps agent (AWS) | budo cost |
| 🟥 | Ch8 Security | security agent + agent security | budo sec |
| ⬛ | Ch9 Capstone | the orchestrated agent team, full TUI | budo v1.0 |