> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/janhq/jan/llms.txt
> Use this file to discover all available pages before exploring further.

# System Requirements

> Hardware and software requirements for running Jan on macOS, Windows, and Linux.

# System Requirements

Jan runs on macOS, Windows, and Linux. Requirements vary based on the size of models you want to run.

## Operating System Requirements

<CardGroup cols={3}>
  <Card title="macOS" icon="apple">
    **Minimum:** macOS 13.6 (Ventura) or later

    **Recommended:** macOS 14+ (Sonoma)

    **Architecture:** Intel or Apple Silicon
  </Card>

  <Card title="Windows" icon="windows">
    **Minimum:** Windows 10 (64-bit)

    **Recommended:** Windows 11

    **GPU Support:** NVIDIA, AMD, Intel Arc
  </Card>

  <Card title="Linux" icon="linux">
    **Supported:** Most distributions

    **Tested:** Ubuntu 18.04+, Debian, Fedora

    **GPU Support:** NVIDIA (CUDA 11.7+)
  </Card>
</CardGroup>

## RAM Requirements by Model Size

Choose models based on your available RAM. For best performance, use models that consume less than 80% of your system RAM.

<Tabs>
  <Tab title="8GB RAM">
    ### Small Models (3B parameters)

    **Model Size:** 2-4GB

    **Examples:**

    * Llama 3B
    * Phi-2
    * Gemma 2B

    **Performance:**

    * Good for basic conversations
    * Lightweight coding assistance
    * Fast response times

    <Warning>
      Avoid models larger than 6GB with 8GB RAM systems.
    </Warning>
  </Tab>

  <Tab title="16GB RAM">
    ### Medium Models (7B parameters)

    **Model Size:** 4-8GB

    **Examples:**

    * Llama 3 8B
    * Mistral 7B
    * Qwen 2.5 7B

    **Performance:**

    * Strong general-purpose capabilities
    * Good coding and reasoning
    * Balanced speed and quality

    <Info>
      Recommended sweet spot for most users.
    </Info>
  </Tab>

  <Tab title="32GB RAM">
    ### Large Models (13B+ parameters)

    **Model Size:** 8-16GB

    **Examples:**

    * Llama 3 13B
    * Mixtral 8x7B
    * Qwen 2.5 14B

    **Performance:**

    * Excellent quality and reasoning
    * Advanced coding capabilities
    * Longer context handling

    Can also run multiple smaller models simultaneously.
  </Tab>

  <Tab title="64GB+ RAM">
    ### Extra Large Models (70B+ parameters)

    **Model Size:** 40GB+

    **Examples:**

    * Llama 3 70B
    * Qwen 2.5 72B

    **Performance:**

    * State-of-the-art quality
    * Complex reasoning tasks
    * Professional use cases

    <Info>
      GPU acceleration highly recommended for these models.
    </Info>
  </Tab>
</Tabs>

## GPU Requirements (Optional)

GPU acceleration significantly improves performance but is not required.

### NVIDIA GPUs

<Tabs>
  <Tab title="Windows">
    **Requirements:**

    * NVIDIA GPU (GTX 10 series or newer recommended)
    * NVIDIA Driver 470.63.01 or higher
    * CUDA Toolkit 11.7 or higher

    **Installation:**

    1. Download [NVIDIA Driver](https://www.nvidia.com/en-us/drivers/)
    2. Install [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads)
    3. Restart your computer
    4. Launch Jan - GPU will be detected automatically

    **Verify Installation:**

    ```bash theme={null}
    nvidia-smi
    nvcc --version
    ```
  </Tab>

  <Tab title="Linux">
    **Requirements:**

    * NVIDIA GPU (GTX 10 series or newer recommended)
    * NVIDIA Driver 470.63.01 or higher
    * CUDA Toolkit 11.7 or higher
    * GCC 11+ with C++17 support

    **Installation:**

    ```bash theme={null}
    # Install NVIDIA driver
    sudo apt update
    sudo apt install nvidia-driver-535

    # Install CUDA toolkit
    # Download from https://developer.nvidia.com/cuda-downloads

    # Install GCC
    sudo apt install gcc-11 g++-11 cpp-11

    # Set up environment
    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64
    ```

    **Verify Installation:**

    ```bash theme={null}
    nvidia-smi
    nvcc --version
    ```
  </Tab>
</Tabs>

### AMD GPUs (Windows)

**Requirements:**

* AMD Radeon GPU (RX 5000 series or newer)
* Latest AMD Adrenalin drivers

**Note:** AMD support is experimental. Performance may vary.

### Intel Arc (Windows)

**Requirements:**

* Intel Arc A-series GPU
* Latest Intel Graphics drivers

**Note:** Intel Arc support is experimental.

### Apple Silicon (macOS)

**Built-in Acceleration:**

* M1, M2, M3, M4 chips have automatic Metal acceleration
* No additional setup required
* Excellent performance for model size

## Verified Configurations

These setups have been tested and confirmed working:

### Desktop Configurations

<Tabs>
  <Tab title="Windows 11 Pro">
    | Component | Specification             |
    | --------- | ------------------------- |
    | OS        | Windows 11 Pro 64-bit     |
    | GPU       | NVIDIA GeForce RTX 4070Ti |
    | CUDA      | 12.2                      |
    | Driver    | 531.18                    |
    | RAM       | 32GB                      |

    **Performance:** Excellent for 13B-70B models
  </Tab>

  <Tab title="Ubuntu 22.04 LTS">
    | Component | Specification             |
    | --------- | ------------------------- |
    | OS        | Ubuntu 22.04 LTS          |
    | GPU       | NVIDIA GeForce RTX 4070Ti |
    | CUDA      | 12.2                      |
    | Driver    | 545                       |

    **Performance:** Excellent for 13B-70B models
  </Tab>

  <Tab title="macOS Sonoma">
    | Component | Specification        |
    | --------- | -------------------- |
    | OS        | macOS 14+ (Sonoma)   |
    | Chip      | M1 Max / M2 Pro / M3 |
    | RAM       | 32GB unified memory  |

    **Performance:** Excellent for 13B-30B models
  </Tab>
</Tabs>

### Virtual Machine Setups

<Warning>
  VM performance is typically lower than bare metal installations.
</Warning>

**Ubuntu on Proxmox:**

| Component | Specification                           |
| --------- | --------------------------------------- |
| OS        | Ubuntu 20.04/18.04 LTS                  |
| VM        | Proxmox                                 |
| GPU       | NVIDIA GeForce GTX 1660Ti (passthrough) |
| CUDA      | 12.1                                    |
| Driver    | 535                                     |

**Requirements for GPU Passthrough:**

* IOMMU support enabled in BIOS
* Proper GPU passthrough configuration
* Dedicated GPU for VM

## Storage Requirements

### Application Size

* **Jan Installation:** 500MB - 1GB
* **System Dependencies:** 2-5GB (CUDA, drivers, etc.)

### Model Storage

Plan disk space based on models you want:

| Model Size | Disk Space Needed |
| ---------- | ----------------- |
| 3B models  | 2-4GB per model   |
| 7B models  | 4-8GB per model   |
| 13B models | 8-16GB per model  |
| 70B models | 40-80GB per model |

**Recommended Free Space:** 50GB+ for comfortable model collection

**Storage Location:**

* **macOS:** `~/Library/Application Support/Jan`
* **Windows:** `C:\Users\%USERNAME%\AppData\Roaming\Jan`
* **Linux:** `~/.config/Jan`

## Network Requirements

### For Local Models

* **Internet:** Only needed to download models initially
* **Offline Use:** Fully functional once models are downloaded

### For Cloud Integrations

* **Internet:** Required for OpenAI, Anthropic, Groq connections
* **Bandwidth:** Standard broadband sufficient

## Performance Optimization

<CardGroup cols={2}>
  <Card title="CPU-Only Systems" icon="microchip">
    * Use smaller models (3B-7B)
    * Increase CPU threads in settings
    * Close other applications
    * Use SSDs for model storage
  </Card>

  <Card title="GPU-Accelerated Systems" icon="bolt">
    * Adjust GPU layers (ngl) parameter
    * Start low and increase gradually
    * Monitor VRAM usage
    * Keep 1-2GB VRAM free for system
  </Card>
</CardGroup>

## Compatibility Notes

### Laptop Considerations

* Laptop GPUs may have lower performance than desktop equivalents
* Thermal throttling can affect sustained performance
* Hybrid graphics (Optimus) may require additional configuration

### Virtual RAM / Swap

* Not recommended as primary RAM for models
* Significantly slower than physical RAM
* May cause poor performance

## Quick Reference

<Info>
  **Most users:** 16GB RAM + 7B models = great experience

  **Budget systems:** 8GB RAM + 3B models = functional

  **Power users:** 32GB+ RAM + GPU + 13B-70B models = excellent
</Info>

## Check Your System

Before installing:

1. **Check RAM:**
   * Windows: Task Manager → Performance → Memory
   * macOS: About This Mac → Memory
   * Linux: `free -h`

2. **Check GPU (if applicable):**
   * Windows: Task Manager → Performance → GPU
   * macOS: Apple Silicon includes integrated GPU
   * Linux: `lspci | grep -i vga`

3. **Check Storage:**
   * Ensure 50GB+ free space for models
   * SSD recommended for best performance

## Need Help?

If you're unsure about compatibility:

* Ask in [Discord #🆘|jan-help](https://discord.gg/FTk2MvZwJH)
* Check [Troubleshooting](/help/troubleshooting) guide
* Review [FAQ](/help/faq) for common questions