How to Start Using OpenCode

After work you want an open-source coding agent in the terminal you already use. OpenCode (project: anomalyco/opencode) is an AI coding agent with a terminal UI, plus desktop and IDE options. This guide stays on the TUI path so you can finish one reviewed Python change in a single evening.

What OpenCode is

Official docs call OpenCode an open source AI coding agent for terminal, desktop, or IDE. In the terminal you install the CLI, connect a model provider, cd into a project, run opencode, optionally /init for an AGENTS.md, then chat. Plan mode (Tab) disables edits so you can agree on a plan before Build mode applies changes. /undo and /redo help when a refactor goes sideways.

You bring your own LLM provider API keys (or OpenCode Zen via /connect). Pick a modern terminal emulator if your default shell UI struggles with TUIs — docs mention options such as WezTerm, Alacritty, Ghostty, and Kitty.

Who it is for

After-work learners who like open source, will paste an API key once, and want agent edits inside a folder of practice .py files. Prefer a hosted IDE agent if you refuse terminal UIs tonight. Prefer OpenCode when you want a local TUI with provider choice and clear plan/build control.

Start steps (one evening)

1. Install. Easiest per docs: curl -fsSL https://opencode.ai/install | bash. Alternatives include npm install -g opencode-ai, Homebrew tap brew install anomalyco/tap/opencode (recommended over the slower-updating core formula), Scoop/Chocolatey on Windows, Mise, or Docker ghcr.io/anomalyco/opencode. Confirm opencode is on your PATH.
2. Connect a provider. Start OpenCode and run /connect. New users can select opencode / OpenCode Zen, sign in at opencode.ai/auth, and paste the API key — or pick another provider you already pay for.
3. Open a tiny project. cd /path/to/practice then opencode.
4. Optional project init. /init analyzes the repo and writes AGENTS.md so later sessions share structure context. Fine on night one for a small folder; skip if you only have one file.
5. Ask, then edit with a bound. Use Tab to enter Plan mode for anything non-trivial; switch back to Build when you accept the plan. For a five-line change, a direct Build ask is enough — still review the diff.

First tiny exercise (15–20 minutes)

Create numbers.py:

def mean(values):
    return sum(values) / len(values)

if __name__ == "__main__":
    print(mean([2, 4, 6]))

In that folder run opencode. Ask (Plan mode first if you want): Add a median for a non-empty list, call it from main after mean, keep under 40 lines.

Approve/build, then:

python numbers.py

If the result is wrong, /undo, tighten the prompt, try again. Quit once mean and median both print — do not start a multi-file rewrite at midnight.

Common pitfalls

• No provider configured. Install alone cannot call a model; finish /connect (or equivalent provider config) before debugging “silent” failures.
• Stale Homebrew core formula. Docs recommend the anomalyco/tap/opencode tap for fresher releases.
• Skipping Plan mode on large asks. Tab to Plan, iterate, then Build — especially when tired.
• Forgetting /undo. One bad refactor should not force a manual git archaeology session on night one (still use git for real projects).
• Sharing sessions accidentally. /share exists for team links; only use it when you intend to publish a transcript.

Next: Python under the agent

OpenCode is fast at proposing code. You still need Python literacy to accept or reject it. Introduction to Python Programming is designed for evening study. Check the course page for the current enrollment window.

Tonight: install, /connect, one small file, one successful python run. Close the TUI on that win.