LM Studio for Home Assistant

Integrate local large language models from LM Studio with Home Assistant. Chat with your LLM through the Home Assistant UI, voice assistants, or any conversation client — and have it query states, control devices, and call services in your smart home.

• Conversation Agents — Full integration with the Home Assistant conversation platform. Chat via UI, Assist, or any conversation client with real-time streaming responses.
• AI Tasks — Generate structured data from natural language prompts using the AI Task platform (JSON schema enforcement, attachments).
• Smart Home Tool Calling — The LLM can automatically query entity states, call services, and control devices through built-in function calling. Multi-turn tool loops are handled transparently.
• Model Management — Automatic on-demand loading/unloading of models with configurable idle timeout to conserve server RAM.
• Multi-Agent Support — Configure multiple conversation agents and AI tasks against a single LM Studio server, each with independent model, settings, and behavior.
• Built-in Utility Tools — Common tools like get_current_time, get_current_date, and get_live_context are available to the LLM without requiring Home Assistant API access.
• MCP Support via LM Studio — LM Studio natively supports the Model Context Protocol (MCP). Configure MCP servers inside LM Studio — the integration automatically passes MCP tools through to the LLM.

• Home Assistant 2025.x or later (requires conversation and ai_task platforms)
• LM Studio running locally or remotely with the local HTTP server enabled (localhost:1234 by default)
• At least one model loaded or available in LM Studio
• (Optional) MCP servers configured in LM Studio for extended tool access

1. Open HACS → Integrations → Three dots (...) → Custom repositories
2. Enter this repository URL and select category Integration
3. Install LM Studio
4. Restart Home Assistant

Copy the custom_components/lmstudio directory into your Home Assistant config:

<config>/custom_components/lmstudio/

Then restart Home Assistant.

All configuration is done through the Home Assistant UI — no YAML required.

Navigate to Settings → Devices & services → Add Integration → search for "LM Studio".

After connecting to the server, add conversation agents or AI tasks as sub-entries:

Same options as conversation agent, excluding System Prompt and HA API Access.

Once configured, you can interact with your agent in several ways:

1. Home Assistant UI — Use the chat panel or Assist interface
2. Voice Assistants — Pipe through any voice assistant that uses the conversation platform
3. Services — Call conversation.process with the agent's entity ID

service: conversation.process
data:
  text: "What's the temperature in the living room?"
target:
  entity_id: conversation.lmstudio_agent

AI tasks generate structured data from natural language prompts. Use them via the ai_task.run service or through automations:

service: ai_task.run
data:
  prompt: "Create a weekly meal plan based on these ingredients: chicken, rice, broccoli"
target:
  entity_id: ai_task.lmstudio_meal_planner

When HA API Access is set to Read or Control, the LLM can automatically:

• Query current states of entities (llm_get_hass_states)
• Call Home Assistant services (llm_call_hass_services)
• Control devices like lights, switches, etc. (llm_control_hass_devices)

The integration handles multi-turn tool loops internally — if the model needs to call tools, it gets results back and can make further decisions until its task is complete.

The following tools are always available to the LLM — no Home Assistant API access needed:

When controlling devices (e.g., "turn on the study light"), the LLM must know the exact entity ID. Use get_live_context(domain="light") to list all lights with their friendly names. This allows the model to find entities even when the entity ID doesn't match the spoken name (e.g., light.study_desk vs "study light").

Config Entry (LM Studio server URL + API key)
    |
    +-- Shared Runtime: {client, model_manager}
    |
    +-- Conversation Agent(s)          +-- AI Task(s)
            |                                |
    conversation.ConversationEntity   ai_task.AITaskEntity
            |                                |
    +-- LmStudioBaseLLMEntity (shared)        |
    +-- Model loading/unloading               |
    +-- Streaming responses                   |
    +-- Tool calling loop                     |

• Single connection, multiple agents — One LM Studio server connection serves all configured sub-entries
• Lazy model loading — Models load on-demand and unload after idle timeout to conserve RAM
• Streaming support — Real-time token streaming for both conversation and AI task responses
• Automatic tool loops — Multi-turn function calling handled transparently

LM Studio supports internet search through its built-in web search feature. Configure a web search provider (e.g., SearXNG, Bing, Google) in LM Studio's settings — the integration passes search tools through automatically to the LLM when available.

This integration uses the official lmstudio Python client library for OpenAI-compatible chat completions. The LM Studio server provides a standard OpenAI-compatible API at /v1/.

MIT © 2025