A local, open-source rival to Claude Code: hands-on setup and first impressions

Goose, an open-source agent framework from Jack Dorsey's Block, combined with the coding-focused Qwen3-coder model, promises a completely free alternative to paid offerings like Claude Code. This is the first of three pieces that walks through getting Goose, the Ollama LLM server, and Qwen3-coder running locally and shares early results from a Mac-based test setup.

Setup is straightforward but requires a few key steps: install Ollama first, choose qwen3-coder:30b inside the app, and trigger a prompt so the 17GB model downloads. Make Ollama visible to the network and set the context length (I used 32K), then install Goose and configure it to connect to Ollama, selecting qwen3-coder:30b as the model.

In use, the stack showed promise but also some rough edges. A simple WordPress plugin test required five rounds of fixes before the agent produced working code; the iterative nature of agentic tools does mean repeated attempts can improve the actual source.

goose, qwen3-coder, ollama, claude code, open-source, jack dorsey, block, local setup, 30b model, wordpress plugin