OpenCode setup: free models and local coding with Ollama

This Windows tutorial installs OpenCode 2.0.27 with Node.js and npm, then uses it for coding tasks. The speaker removes the older opencode-ai package because it shares the same command with the demonstrated installation.

The default model fails because it has been deprecated. The speaker selects Big Pickle through /models and reports that four free cloud options answered without an account. Availability can change, and cloud requests send code to servers.

Shift+Tab switches between Plan and Build. Planning leaves the project folder empty; Build then writes a single-file to-do page in about 23 seconds. The speaker checks that tasks survive a refresh. /init creates AGENTS.md project notes, and a fresh session adds filters and a dark mode setting that persists after refresh.

The local section selects Qwen 3.8 through Ollama and asks it to write and run a Python tip calculator. The speaker reports about 38 seconds on an RTX 5090, including model loading, and checks ollama ps for GPU use. The source description identifies a 27B model using about 17 GB of graphics memory. These are results from the speaker's setup. He warns that local coding requests can be slow on a CPU and distinguishes local processing from the cloud route.