opencode

opencode is an open-source coding agent in the same shape as Claude Code : a terminal agent that reads, edits, and runs code in a repository. Its useful property here is that you can point it at any model endpoint, including a local one. So in this stack opencode is the local lane — it drives the open model you serve with Ollama for the routine 80% of the work (see local and frontier models ).

Install and point it at a local model

Install opencode from its project page, then set its model endpoint to your local server — for example Ollama at http://<gpu-machine>:11434. The exact config lives in opencode’s own settings; the key is that the base URL is your local endpoint, not a hosted API.

amux runs opencode in sessions the same way it runs Claude Code, so the local lane gets the same isolated worktrees and dashboard.

Local models only on this lane

Point opencode at local models — that is what it is for here. Wiring it to a paid subscription’s backend, to use it as if that were an open API, generally breaks the provider’s terms. Keep opencode on local models, and use the subscription’s own client (Claude Code) for the frontier lane.

Serve the model it talks to: Ollama .
Why API keys and local models, not a subscription: subscriptions, APIs, and local models .
The lane split: local and frontier models .

Source: content/tools/opencode.md · maintained in the nuilab-agenticai repository.

Install and point it at a local model

Next