Skip to content

opencode_session_one.md

Latest commit

 

History

History
206 lines (132 loc) · 6.06 KB

opencode_session_one.md

File metadata and controls

206 lines (132 loc) · 6.06 KB
marp true
theme codebar
paginate true
title opencode — Session One
description Intro to opencode and free models via opencode zen
author codebar
header AI Workshop · opencode
footer codebar · AI Workshop
html true
codebar · AI Workshop · Session 01

opencode

Agentic coding with open-source models

Presenter: Karel M. - @oricha Co-host: Andy D. @mkdir-andy

Sponsor: Factoria F5

Agenda

  1. opencode introduction
  2. Local setup & tool tour
  3. Live demo
  4. Hands-on with coaches
  5. Share & discuss

How we code is changing

Five years ago, our job was to type code. Today, increasingly, it is to steer it.

  • Editors → autocomplete → chat → agents that act.
  • The bottleneck moves from writing to reviewing and deciding.
  • The gate used to be the subscription. Today there are free, open-source paths in.

Today is about picking one up and trying it.

What is a code agent?

An agent is an LLM in a loop with tools.

  • It reads your files, runs commands, writes code, observes results, decides what to do next.
  • The LLM inside is the model — Claude, GPT, Big Pickle, DeepSeek…
  • Each model has its own strengths, cost, and access requirements.

The next choice is which model — and therefore which provider — does the work.

Free vs paid — closer than you'd think

Share of real GitHub issues each model resolves on SWE-bench Verified — free models are within striking distance.

Claude Opus 4.5 paid
82%
Claude Sonnet 4.5 paid
77%
GPT-5 paid
74%
DeepSeek V4 Flash free
65%
Nemotron 3 Ultra free
62%
Big Pickle free
58%
MiMo V2.5 free
52%
North Mini Code free
48%
Illustrative — check the live leaderboard at swebench.com.

Why opencode

  • Open source — community-driven, self-hostable, no vendor lock-in.
  • 75+ providers — Anthropic, OpenAI, Google, local models, opencode zen.
  • Free tier via opencode zen — try real agentic coding without a credit card.
  • TUI-first — built for the terminal; pairs well with WezTerm, Alacritty, Ghostty, Kitty.
  • Composable — define custom agents and project rules in AGENTS.md.

TUI tour

Three regions to know:

  • Input area (bottom) — your prompts and slash commands.
  • Transcript (centre) — the agent's actions, file diffs, tool output.
  • Status bar (top/bottom) — current agent · current model · session name.

Slash commands you will use first: /init, /connect, /models, /share.

opencode TUI

Interacting — agents, models, sessions

What Key What it does
Agents tab / shift+tab · ctrl+x a Switch the persona driving the session
Models f2 / shift+f2 · ctrl+x m Change which model is doing the work
Sessions ctrl+x l · ctrl+x g · ctrl+r List, open timeline, rename

Full reference: setup/commands.md.

Live demo

Build a small project, live, using opencode + a free zen model.

While you watch, pay attention to:

  • Prompt style — intent-first, not step-by-step.
  • When to switch agents — research vs. implement vs. review.
  • Undo / redo — /undo and /redo are your safety net.

Hands-on with coaches

Each student is paired 1:1 with a coach — together you pick one challenge and work through it as a team.

  1. Challenge one — context: TBD
  2. Challenge two — context: TBD
  3. Challenge three — context: TBD

Before you start: fork github.com/codebar/ai-workshop-opencode and clone your fork locally — that gives you a sandbox to commit your work into.

Share & discuss

Last 15 minutes — open mic.

  • Show one thing the agent did well.
  • Show one thing it got wrong and how you steered it.
  • Ask the room: prompts, models, workflows, surprises.
Thank you

Keep going