OpenCode tutorial: permissions, plugins and Ollama setup

OpenCode is an open source terminal coding assistant that the presenter explores by building a Flask to-do application. He prefers its interface to Claude Code and Codex, but frames that preference as personal taste rather than proof of better performance. The tutorial starts with OpenCode already installed and does not cover installation.

The demonstration focuses on controlling changes. After the agent initially edits files and runs commands without approval, the presenter configures edit and shell permissions to ask first, with an exception for plain ls. Adding flags triggers an approval prompt. He then requests SQLite3 persistence and opens a full-screen diff before accepting changes.

Other examples cover undo and redo, session switching, external prompt editing in Neovim, and read-only plan mode before implementation. Session forks share the working files; they do not create separate code branches. An attempt to clear a prompt through the external editor fails during the demonstration.

The presenter also configures OpenCode Notifier and receives a completion notification. For local AI use, he explains an Ollama provider configuration with an OpenAI-compatible package, a LAN endpoint and a 20-billion-parameter GPT model. This final section uses a screenshot rather than a live inference test because of his laptop's limited hardware.