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

# Local Models

> Run AI models on-device for maximum privacy with LM Studio or Ollama

Parsaa's **Local** provider connects to any OpenAI-compatible server running on your Mac, so inference happens on-device with no network calls. Your code never leaves your machine.

<Info>
  The Local provider is configured under **Settings → Models → Custom Providers**. No API key is required — you only point Parsaa at the local endpoint. The default base URL is `http://localhost:1234/v1`.
</Info>

## LM Studio

<Steps>
  <Step title="Install LM Studio">
    Download LM Studio from [lmstudio.ai](https://lmstudio.ai).
  </Step>

  <Step title="Download a model">
    Browse the catalog and download a coding-focused model.
  </Step>

  <Step title="Start the local server">
    Open the **Local Server** tab and start it. LM Studio serves an OpenAI-compatible API (default `http://localhost:1234/v1`).
  </Step>

  <Step title="Configure Parsaa">
    In **Settings → Models → Custom Providers**, select the **Local** provider and confirm the base URL matches LM Studio.
  </Step>

  <Step title="Select the model">
    Choose the local model from the picker. All inference runs on-device.
  </Step>
</Steps>

## Ollama

<Steps>
  <Step title="Install Ollama">
    Download Ollama from [ollama.com](https://ollama.com).
  </Step>

  <Step title="Pull a model">
    ```bash theme={null}
    ollama pull qwen2.5-coder
    ```
  </Step>

  <Step title="Confirm the endpoint">
    Ollama exposes an OpenAI-compatible API at `http://localhost:11434/v1`.
  </Step>

  <Step title="Configure Parsaa">
    In **Settings → Models → Custom Providers**, select the **Local** provider and set the base URL to Ollama's endpoint.
  </Step>

  <Step title="Select the model">
    Choose your local model from the picker.
  </Step>
</Steps>

## Hardware Recommendations

<Info>
  **Apple Silicon with 16GB+ RAM** is recommended. Larger models (13B+) benefit from 32GB+. Apple's unified memory makes M-series chips well-suited to local inference.
</Info>

| Model Size      | Minimum RAM | Recommended RAM |
| --------------- | ----------- | --------------- |
| 7B parameters   | 8 GB        | 16 GB           |
| 13B parameters  | 16 GB       | 32 GB           |
| 34B+ parameters | 32 GB       | 64 GB           |

## Privacy

With a local model, requests never touch the network — everything runs on your Mac. This suits:

* **Proprietary codebases** that can't leave the organization
* **Regulated industries** with data-residency requirements
* **Air-gapped environments** with no internet access

<Note>
  Local models trade some capability for privacy. Frontier cloud models generally produce higher-quality results on complex tasks. Choose based on your privacy needs and the difficulty of the work.
</Note>