Summary
A complete setup guide for running Qwen 2.5 Coder locally via Ollama and connecting it to OpenCode, creating a private, offline-capable AI coding assistant in your terminal.
Key Technical Findings
- Setup involves installing Ollama and OpenCode on macOS, Linux, or Windows (WSL2).
- Requires at least 8 GB of RAM (16 GB recommended for best performance).
- Qwen 2.5 Coder model is pulled from Ollama with a command.
- Configuration of OpenCode to use Qwen locally involves editing the opencode.json file to set num_ctx to 32768 for better context handling.
Commands Used
`curl -fsSL https://ollama.com/install.sh | sh` (for macOS/Linux)
`brew install ollama` (macOS Homebrew method)
`ollama --version` (to verify Ollama installation)
`ollama pull qwen2.5-coder` (to download the Qwen 2.5 Coder model)
`ollama list` (to confirm model is available)
`npm install -g opencode-ai` (to install OpenCode via npm)
`brew install sst/tap/opencode` (macOS Homebrew method for OpenCode)
`opencode --version` (to verify OpenCode installation)
Edit `~/.config/opencode/opencode.json` as per provided config
`ollama serve` (to start Ollama server)
`cd ~/projects/my-project && opencode` (to run OpenCode in a project directory)
Lessons Learned
- Increasing num_ctx to 32768 improves context handling for coding tasks.
- Tool calls may not work if the context window is too small; increasing num_ctx can resolve this.