opencode-local-models Plugin

Auto-discovers Ollama and LM Studio chat models in opencode at startup. No hardcoded model lists to keep in sync.

Why this plugin?

opencode does not discover local models for you. To use Ollama or LM Studio, you add a provider entry and then hand-write the models map, one entry per model. Every time you pull a model, remove one, or change a tag, the config goes stale.

opencode also merges in preset models from models.dev for a matching provider id. So the /models picker can show entries that your engine does not actually serve. Other plugins exist, but they usually cover one engine, generate config files, or need a separate sync command.

This plugin does three things differently:

• It asks the live engine what is running at startup and fills the model list from that answer.
• It covers both Ollama and LM Studio in one plugin.
• If an engine has no chat models, the provider stays empty and does not show in the picker at all. You only ever see live, usable chat models.

Related reading: opencode providers docs

Requirements

• opencode 1.18.19 or newer
• Ollama running on http://localhost:11434 (optional)
• LM Studio 0.4.21 or newer with the local server enabled (optional)

The plugin reads LM Studio's native API at GET /api/v1/models. That endpoint reports each model's type and context. The OpenAI-compatible /v1/models endpoint does not report those fields. Any engine can be absent. The plugin simply skips it.

Quick start

Install the plugin into opencode's global plugins directory:

curl -fsSL https://raw.githubusercontent.com/nisrulz/opencode-local-models/v0.1.1/install.sh | bash

The script copies the plugin to ~/.config/opencode/plugins/opencode-local-models.js and adds the ollama and lmstudio providers to ~/.config/opencode/opencode.json.

opencode auto-loads every file in the plugins directory when it starts, so the plugin needs no extra config. For a project-level or manual install, see the install instructions.

Restart opencode, run /models, and search for ollama or lmstudio. Once an engine is running with a chat model loaded, you will see live entries like ollama/lfm2.5:8b.

Models are discovered when opencode starts, not while it runs. If you add or remove models on an engine, restart opencode so the picker catches up.

Docs

• Options
• Under the hood
• Development

License

Apache License, Version 2.0