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

# LM Studio

> Run local models with LM Studio in EnConvo

## Overview

LM Studio is a desktop application for running large language models locally on your Mac. It provides a user-friendly GUI for discovering, downloading, and running GGUF models with no API key required. Like Ollama, it keeps all data on your machine for complete privacy.

## Supported Models

LM Studio supports any GGUF-format model. Popular choices include:

| Model             | Size    | Best For          |
| ----------------- | ------- | ----------------- |
| **Llama 3.1 8B**  | \~5 GB  | General purpose   |
| **Llama 3.1 70B** | \~40 GB | Complex tasks     |
| **Mistral 7B**    | \~4 GB  | Fast responses    |
| **Phi-3 Mini**    | \~2 GB  | Lightweight tasks |
| **CodeLlama 7B**  | \~4 GB  | Programming       |
| **Qwen 2.5 7B**   | \~5 GB  | Multilingual      |

## Setup

<Steps>
  <Step title="Install LM Studio">
    1. Download from [lmstudio.ai](https://lmstudio.ai)
    2. Install the application on your Mac
    3. Launch LM Studio
  </Step>

  <Step title="Download a Model">
    1. In LM Studio, go to the **Discover** tab
    2. Search for a model (e.g., "Llama 3.1")
    3. Click **Download** on your preferred quantization (Q4\_K\_M recommended for balance)
    4. Wait for the download to complete
  </Step>

  <Step title="Start the Local Server">
    1. Go to the **Developer** tab in LM Studio
    2. Select your downloaded model
    3. Click **Start Server**
    4. Note the server address (default: `http://localhost:1234`)
  </Step>

  <Step title="Configure in EnConvo">
    1. Open **Settings** → **AI Provider**
    2. Select **LM Studio**
    3. Go to **Credentials** module
    4. Set the endpoint to `http://localhost:1234` (or your custom port)
  </Step>

  <Step title="Select Model">
    Choose from your loaded models in the dropdown
  </Step>
</Steps>

<Tip>
  No API key is needed -- LM Studio runs entirely on your local machine.
</Tip>

## Configuration

| Setting         | Description            | Default                 |
| --------------- | ---------------------- | ----------------------- |
| **Endpoint**    | Local server address   | `http://localhost:1234` |
| **Model Name**  | Currently loaded model | Auto-detected           |
| **Temperature** | Creativity (0-2)       | Medium (1)              |

## Validate and Use

<Steps>
  <Step title="Start the server first">
    In LM Studio's **Developer** tab, load a model and start the local server before validating in EnConvo.
  </Step>

  <Step title="Validate the endpoint">
    Use `http://localhost:1234` unless you changed the LM Studio port. EnConvo expects the server base URL, not a full `/chat/completions` path.
  </Step>

  <Step title="Keep LM Studio open">
    LM Studio must keep the server running while EnConvo uses it. If the app is closed or the model is unloaded, requests will fail.
  </Step>
</Steps>

## System Requirements

| RAM    | Recommended Models          |
| ------ | --------------------------- |
| 8 GB   | 7B models (Q4 quantization) |
| 16 GB  | Larger 7B/13B models        |
| 32 GB  | 30B models                  |
| 64 GB+ | 70B models                  |

<Note>
  Apple Silicon Macs (M1/M2/M3/M4) with Metal acceleration provide significantly better performance than Intel Macs for local LLM inference. GPU VRAM (unified memory) is the key factor for model size.
</Note>

## LM Studio vs Ollama

| Feature             | LM Studio         | Ollama                 |
| ------------------- | ----------------- | ---------------------- |
| **Interface**       | GUI application   | Command-line           |
| **Model format**    | GGUF              | GGUF (auto-managed)    |
| **Model discovery** | In-app browser    | `ollama pull` command  |
| **Server control**  | Manual start/stop | Auto-starts on install |
| **Configuration**   | Visual settings   | Config files           |

## Privacy Benefits

<CardGroup cols={2}>
  <Card title="Complete Privacy" icon="shield-halved">
    All data stays on your Mac
  </Card>

  <Card title="Offline Access" icon="wifi-slash">
    Works without internet after model download
  </Card>

  <Card title="No Usage Limits" icon="infinity">
    Run unlimited queries locally
  </Card>

  <Card title="No Cost" icon="dollar-sign">
    Free to use, no API fees
  </Card>
</CardGroup>

## Troubleshooting

<AccordionGroup>
  <Accordion title="Connection refused">
    * Ensure LM Studio's local server is running (check the Developer tab)
    * Verify the port matches your EnConvo configuration (default: 1234)
    * Check that no other application is using the same port
  </Accordion>

  <Accordion title="Slow responses">
    * Use a smaller model or higher quantization (Q4\_K\_M or Q4\_K\_S)
    * Close other memory-intensive applications
    * Check that Metal GPU acceleration is enabled in LM Studio settings
  </Accordion>

  <Accordion title="Model fails to load">
    * Ensure you have enough RAM for the model size
    * Try a smaller quantization variant
    * Re-download the model if the file may be corrupted
    * Restart LM Studio and try again
  </Accordion>

  <Accordion title="No models in dropdown">
    * Make sure a model is loaded and the server is started in LM Studio
    * Refresh the model list in EnConvo settings
    * Check the endpoint URL is correct
  </Accordion>
</AccordionGroup>