Local Model Squad (Ollama / LM Studio)

A lead and a helper that run entirely on a model server on your own machine or network: no cloud model, no per-token bill.

What's inside

• Lead (OpenCode) and Helper (Qwen Code), both set to the custom provider you choose at import and the model qwen3-coder:30b. The lead hands small, self-contained subtasks to the helper over its arrow.
• Shell terminal the lead runs commands and tests in.
• Local setup note with the checklist below.

Before you import

1. Run an OpenAI-compatible server: Ollama, LM Studio, llama.cpp or vLLM.
2. Add it in NeuroSquad: Settings, Custom providers. The app tests which APIs it speaks; both cards need the Chat Completions endpoint.
3. Pull a coding model with tool calling. The cards ask for qwen3-coder:30b (Ollama's name); with another model or LM Studio's naming, change the model on each card after import.

How to use

At import, pick your provider for "Local model server". Without one, the cards start on their CLIs' own logins instead. Then describe the task in the Lead's pasted prompt.