The OpenCode course

This is a hands-on course, not a reference. You’ll spend a week with OpenCode turning an inherited side project — feedmill, a self-hosted, terminal-first feed aggregator — into something you trust running unattended. It’s a small Go CLI plus an HTTP sync server that pulls dozens of RSS, Atom, and JSON feeds, dedupes them, and serves a reading queue you read from the terminal. It mostly works, it’s thin on docs, and you already suspect one feed type parses its dates wrong. Every lesson takes a problem you’ll really hit and shows the exact OpenCode move that fixes it.

This course is differentiator-first. OpenCode isn’t tied to one vendor’s model, it lives in a full-screen terminal TUI instead of a chat sidebar, and it’s an open client/server you can drive headless and share by link. So we lead with the things no closed tool can match — bring-your-own-model across 75+ providers, the keyboard-driven TUI, the share-and-serve architecture — before settling into the agent primitives every tool shares (AGENTS.md, skills, subagents, MCP/LSP). If you want the spec-level “what is this primitive” view, that’s Foundations. This course is the other half: the muscle memory.

Already use another CLI? The Codex and Claude Code courses are its siblings — same teach-by-doing spine, a different tool. OpenCode shares the AGENTS.md standard with Codex, so that chapter will feel familiar.

All chapters live

The full curriculum is written start to finish. Read it in order to build up from zero, or jump straight to the chapter you need. Lessons may still get sharpened over time.

The curriculum

Ten chapters, in reading order. Each is a self-contained chapter that carries the feedmill project one stretch further — start at the top, or pick the problem in front of you.

1 · Getting started Install OpenCode, bring your own model, watch one read-only turn of the loop, ship a first reviewed fix, and dock it next to your editor — where it reads the same LSP diagnostics your editor draws as red squiggles.

2 · Providers & models OpenCode's signature move: wire up multiple providers, switch models mid-session, set a per-agent model, run a task against two providers side by side, point at a local model, and lean on Zen — OpenCode's own curated list of tested, coding-grade models.

3 · Living in the TUI Get fluent in the keyboard-first TUI and make it your continuity layer: the leader key and palette, listing, resuming and forking sessions, undo/redo and compaction that runs automatically when context fills (or on demand with /compact), Tab between plan and build, custom keybinds and themes.

4 · Rules (AGENTS.md) Stop the agent reintroducing the same feedmill bugs. Write project and user rules in AGENTS.md — the open standard OpenCode shares with Codex — learn its discovery walk, the nesting trap, and the @path-import sharp edge.

5 · Skills Package the onboard-a-new-feed dance once as a SKILL.md the agent invokes itself — and reuse skills written for Claude Code or stored in generic .agents paths, since OpenCode reads all of them.

6 · Subagents Push noisy scans into isolated context and fan parallel audits across workers. Use the three built-in subagents — General, Explore, and Scout — by @mention, or author your own with their own model, permissions, and tools.

7 · Permissions & isolation Set the leash per tool per agent, and switch the whole policy by switching primary agents. Honest about the sharp edge: as of v1.16.2 OpenCode ships no documented built-in OS sandbox — its permission system is a UX guardrail, not a security boundary, so real isolation means running the CLI in a container.

8 · Extending OpenCode Reach beyond the repo: MCP servers in the mcp field, the LSP field as the in-tree code-intelligence channel the agent gets for free, and plugins — live JS/TS modules that subscribe to events and register tools.

9 · Sharing & headless Two distinctive moves: /share turns a debugging session into a publicly accessible URL for a second opinion (mind what's in it), and the client/server split lets you drive feedmill headless with `opencode run --format json`, in CI, against its own server.

10 · Daily workflow The habits that compound: the custom slash commands you wrote, the prompting and keyboard reflexes that get a clean first pass, and a look back across the differentiator-first arc.

How to read it

Each chapter carries one running project — you’ll follow feedmill across a week of real work — and the lessons build on each other: each one picks up a problem the last one left you with, adds a move to your repertoire, and hands off to the next. The read-before-edit, commit-at-green discipline you set up in chapter one is the spine every later chapter leans on, and LSP runs as a quiet through-line — the skeptic’s hook in chapter one, the in-tree code-intelligence channel again in Extending.

So you get the most out of a chapter by reading it in order. But each lesson still stands on its own if you drop in from a search with one specific problem in front of you — you’ll just miss the running story.

What you won’t find is a fixed template stamped onto every lesson. Some need a long worked transcript, some a single command and a warning, some a digression into why the tool behaves the way it does. The shape follows the content, the way a good book varies its chapters.