opencode-models-discovery

A universal OpenCode plugin for dynamic model discovery across any OpenAI-compatible provider.

Originally inspired by opencode-lmstudio, this project has been refactored into a general-purpose model discovery plugin with provider-level discovery controls, model filtering, metadata enrichment, and /connect-backed credential support.

Features

• Works with any OpenAI-compatible provider
• Discovers models dynamically from provider model endpoints
• Injects discovered models into OpenCode provider config automatically
• Supports provider-level include, exclude, and endpoint overrides
• Supports regex-based model filtering
• Can enrich model limits and reasoning metadata from provider-specific endpoints
• Supports OpenCode /connect credentials for custom providers

Installation

npm install opencode-models-discovery
# or
bun add opencode-models-discovery

Quick Start

Add the plugin to your opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": [
    "opencode-models-discovery@latest"
  ],
  "provider": {
    "lmstudio": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "LM Studio (local)",
      "options": {
        "baseURL": "http://127.0.0.1:1234/v1",
        "modelsDiscovery": {
          "enabled": true
        }
      }
    }
  }
}

On startup, the plugin will query the provider's models endpoint and merge discovered models into the active OpenCode config.

v0.12 Transition and v1.0 Compatibility

Version 0.12.x is a transition line for the next configuration model. Existing plugin-level discovery options continue to work in 0.12.x, but they are deprecated and will be removed in 1.0.0.

Deprecated plugin-level options:

• discovery.enabled
• providers.include
• providers.exclude
• models.includeRegex
• models.excludeRegex
• smartModelName

Recommended configuration should live on each provider instead:

{
  "provider": {
    "lmstudio": {
      "options": {
        "modelsDiscovery": {
          "enabled": true,
          "models": {
            "includeRegex": "^llama",
            "excludeRegex": "embedding"
          },
          "smartModelName": true,
          "modelInfoFormat": "models.dev"
        }
      }
    }
  }
}

Only add the fields you need. For example, do not add regex filters unless you actually want filtering.

Planned 1.0.0 behavior:

• Plugin-level global discovery config will be removed.
• Discovery will remain enabled by default for providers unless disabled.
• Set provider.<id>.options.modelsDiscovery.enabled = false to disable discovery for one provider.
• A future OPENCODE_MODELS_DISCOVERY_DEFAULT_ENABLED=false environment variable is planned for users who want unspecified providers to default to disabled.

When 0.12.x detects deprecated global config, it logs a warning, shows a migration toast, and injects /models-discovery:migrate into OpenCode commands.

Helper Commands

The plugin injects helper commands into OpenCode's runtime command list.

/models-discovery:config

Opens an assistant-guided configuration flow using OpenCode's customize-opencode skill. Use this when setting up the plugin, adding provider-level discovery config, enabling metadata enrichment, or disabling discovery for a provider.

This command is available whenever the plugin is loaded.

/models-discovery:migrate

Opens an assistant-guided migration flow using OpenCode's customize-opencode skill. It looks for OpenCode config files that declare this plugin and moves deprecated plugin-level discovery options into provider.<id>.options.modelsDiscovery where safe.

This command is injected only when deprecated global discovery config is detected.

The migration assistant is instructed to inspect project config, user global config, and OPENCODE_CONFIG when present. It should not edit managed or organization-controlled config unless you explicitly ask it to.

Model Metadata Enrichment

Discovery adds model ids to your OpenCode provider config. Some providers only expose minimal /models responses, so the plugin can optionally enrich discovered models with OpenCode-compatible capability metadata such as context limits, output limits, reasoning, tool calling, attachments, structured output, temperature support, and modalities.

Metadata enrichment is explicit. The plugin does not contact external metadata sources unless configured.

For models.dev enrichment:

{
  "modelsDiscovery": {
    "enabled": true,
    "modelInfoFormat": "models.dev"
  }
}

For LiteLLM-compatible model info endpoints:

{
  "modelsDiscovery": {
    "enabled": true,
    "modelInfoEndpoint": "/v1/model/info",
    "modelInfoFormat": "litellm"
  }
}

If metadata cannot be fetched or matched safely, discovery still succeeds and the plugin leaves unknown capability fields unset rather than guessing defaults. See docs/providers.md for details.

Upgrade Note

If you upgrade the plugin and OpenCode still behaves like it is using an older build, refresh the OpenCode plugin cache and restart OpenCode.

This can happen because OpenCode may continue using a previously cached package after the npm package itself has been updated.

After changing opencode.json, restart OpenCode. OpenCode loads config at startup, so command and provider changes are not guaranteed to take effect in an already-running session.

/connect Support

For custom OpenAI-compatible providers, you still define the provider in opencode.json so OpenCode and this plugin know the provider id, npm package, and baseURL.

If the API credential is managed through OpenCode /connect, you do not need to duplicate the same key in provider.<name>.options.apiKey.

Discovery auth precedence is:

1. provider.<name>.options.apiKey
2. OpenCode resolved provider key, when available during plugin startup
3. OpenCode /connect auth store for same-id type: "api" credentials

Details and examples: docs/connect-and-auth.md

Mimocode Compatibility

This plugin is also compatible with Mimocode as an OpenCode-compatible host.

When startup-time discovery needs to recover /connect-managed API credentials from the local auth store, the plugin selects the host data directory from runtime environment markers:

• OPENCODE=1 or no host marker: ~/.local/share/opencode/auth.json
• MIMOCODE=1: ~/.local/share/mimocode/auth.json

This keeps the same provider configuration model while allowing the plugin to work in both OpenCode and Mimocode environments.

Documentation

• Configuration guide: docs/configuration.md
• /connect credentials and auth-backed discovery: docs/connect-and-auth.md
• Provider compatibility and detection rules: docs/providers.md
• Upgrade notes: docs/upgrading.md

Requirements

• OpenCode with plugin support
• At least one OpenAI-compatible provider running locally or remotely
• Provider API accessible through either a /v1-style base URL or an explicitly configured discovery endpoint

Logging

When available, the plugin writes logs through OpenCode's structured server log API via client.app.log(...) using the service name opencode-models-discovery.

If structured logging is unavailable in the runtime, the plugin falls back to prefixed console.* output. Key log categories are emitted through metadata such as plugin, config, discovery, event, and filtering to make local debugging easier with opencode --print-logs.

Star History

Contributing

Contributions are welcome. Please feel free to submit a Pull Request.

License

MIT

Disclaimer

This project is not built by the OpenCode team and is not affiliated with OpenCode in any way.

This project is not affiliated with, endorsed by, or sponsored by models.dev. Optional models.dev metadata enrichment uses the public models.dev index only when explicitly configured by the user.