opencode-plugboard

Appearance

Home Assistant: Extended OpenAI Conversation ​

The Extended OpenAI Conversation custom component (not core Home Assistant's OpenAI integration). It adds a Base Url to the OpenAI conversation agent and lets the model call Home Assistant services.

Plugboard options ​

Home Assistant reaches plugboard over the LAN, so plugboard binds a reachable address with keys set (Serving other machines):

jsonc
{
  "options": {
    "listen": { "host": "0.0.0.0", "port": 4010 },
    "keys": { "homeassistant": "{env:PLUGBOARD_KEY_HOMEASSISTANT}" },
    "models": ["anthropic/claude-sonnet-4-6"],
    "aliases": { "assist": "anthropic/claude-sonnet-4-6" },
    "apps": { "homeassistant": { "allowedModels": ["assist"] } }
  }
}

Connect ​

  1. Install the component from HACS (as a custom repository) and restart Home Assistant.
  2. Settings → Devices & Services → Add Integration → Extended OpenAI Conversation: the homeassistant key as API Key, http://192.0.2.10:4010/v1 (your plugboard host) as Base Url, Skip Authentication off.
  3. In the integration's options, set Chat model to assist.
  4. Settings → Voice assistants: pick Extended OpenAI Conversation as the conversation agent.
  5. Expose the entities the model may control.

Setup checks the key and URL with GET /v1/models; a wrong chat model only fails on the first turn.

What works ​

Conversation turns with streaming and service calls (Home Assistant runs them), and AI Task structured output.

Limits ​

  • Raise Max tokens above the default 500 for a model that reasons, or it may leave no visible answer.
  • A custom Functions list replaces the default execute_services tool; copy it into your list to keep it.
  • Don't give an alias a name starting with gpt-5 or o1–o4: the component then changes which parameters it sends.
  • To change the URL or key later, call the extended_openai_conversation.change_config service.