Get Agent Loadout
Coding agentsIntermediateDocs-checked, not yet hands-on tested

Local, private coding agent

Use an AI coding agent on code that must never leave your machine.

Starter prompt

You are running fully offline. Read the repository structure first and summarize it in 10 lines. Then <task>. Make small edits, run the tests after each change, and stop to ask if a change touches more than 3 files.

Config

terminal

ollama pull qwen3-coder:30b
# one-command launcher from the Ollama library page:
ollama launch opencode --model qwen3-coder

Setup steps

  1. Install Ollama and pull qwen3-coder:30b.
  2. Install OpenCode, or use the ollama launch command.
  3. Run the starter prompt inside your repo.
  4. Keep tasks small; local models do best with tight scopes.

Why this works

Nothing is sent to a cloud API, and the 256K context window is large enough for repo-scale work.

Watch out

Local models are slower and weaker than frontier cloud models on hard tasks. Ollama also lists a cloud variant (480b-cloud); don't use it if privacy is the point.

Related loadouts

Compare the gear: AI coding agents tier list