> ## Documentation Index
> Fetch the complete documentation index at: https://docs.spaceduck.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# LM Studio

> Run Spaceduck with LM Studio for a GUI-based local model experience with chat and embeddings.

## What it's good for

The easiest local setup. LM Studio has a built-in model browser, one-click downloads, and runs an OpenAI-compatible server out of the box. Great if you want local models without managing GGUF files and command-line flags.

## Requirements

* [LM Studio](https://lmstudio.ai) installed
* A model downloaded through LM Studio's model browser
* The local server started in LM Studio (Developer tab)

## Configure in Spaceduck

### Chat

<Steps>
  <Step title="Start the LM Studio server">
    1. Open LM Studio
    2. Load a model (e.g., Qwen3, Llama 3.1, DeepSeek)
    3. Go to the **Developer** tab
    4. Click **Start Server** — it runs at `http://localhost:1234` by default
  </Step>

  <Step title="Configure Spaceduck">
    In **Settings > Chat**:

    * **Provider**: LM Studio
    * **Base URL**: `http://localhost:1234/v1`
    * **Model**: the model identifier shown in LM Studio (e.g., `qwen/qwen3-4b-thinking-2507`)

    Or via CLI:

    ```bash theme={null}
    spaceduck config set /ai/provider lmstudio
    spaceduck config set /ai/model "qwen/qwen3-4b-thinking-2507"
    ```

    <Info>
      LM Studio doesn't require an API key by default. Spaceduck sends a dummy key (`lm-studio`) for servers that require the Authorization header to be present.
    </Info>
  </Step>

  <Step title="Verify">
    ```bash theme={null}
    curl http://localhost:1234/v1/models
    ```

    You should see your loaded model listed.
  </Step>
</Steps>

### Embeddings

LM Studio can serve embeddings from the same server, or you can run a second instance on a different port.

<Steps>
  <Step title="Load an embedding model">
    In LM Studio, load an embedding model alongside your chat model, or start a second server instance on a different port.
  </Step>

  <Step title="Configure Spaceduck">
    In **Settings > Memory**:

    * Toggle **Semantic recall** on
    * **Provider**: LM Studio
    * **Server URL**: `http://localhost:1234/v1` (same server, or a different port if running separately)
    * **Model**: your embedding model identifier
    * **Dimensions**: match the model (e.g., 768 for nomic, 1024 for large models)

    Or via CLI:

    ```bash theme={null}
    spaceduck config set /embedding/enabled true
    spaceduck config set /embedding/provider lmstudio
    spaceduck config set /embedding/model "nomic-ai/nomic-embed-text-v1.5"
    spaceduck config set /embedding/dimensions 768
    ```
  </Step>

  <Step title="Test">
    Click the **Test** button in **Settings > Memory** to verify the embedding connection.
  </Step>
</Steps>

## Test and troubleshoot

| Problem                     | Cause                             | Fix                                                      |
| --------------------------- | --------------------------------- | -------------------------------------------------------- |
| `ECONNREFUSED` on port 1234 | LM Studio server not started      | Go to Developer tab and click Start Server               |
| Model not found             | Model identifier doesn't match    | Check `curl localhost:1234/v1/models` for the exact name |
| Slow responses              | Model too large for available RAM | Try a smaller model or lower quantization                |
| `<think>` tags in output    | Thinking model (Qwen3, DeepSeek)  | Spaceduck strips these automatically — this is expected  |

<Tip>
  LM Studio's model browser makes it easy to try different models. Download a few, test them with Spaceduck, and keep the one that works best for your use case.
</Tip>