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

# Ollama (Local Models)

> Run open-source models locally with zero data egress using Ollama.

<Snippet file="ollama-note.mdx" />

Ollama lets you run open-source LLMs entirely on your machine. No API key required — your data never leaves your computer. Both the **RLM** and **Native Agent** engines support Ollama.

## Prerequisites

* **macOS/Linux:** Install Ollama from [ollama.ai](https://ollama.ai)
* **Windows:** Ollama is available as a preview for Windows — see [ollama.ai/download](https://ollama.ai/download)
* Minimum 8 GB RAM for 7B models; 16 GB recommended for 13B+ models

## Install Ollama and pull a model

```bash theme={null}
# Install (macOS with Homebrew)
brew install ollama

# Pull any model you want to use
ollama pull llama3.1
ollama pull mistral
ollama pull qwen2.5:14b     # 14B — better quality, needs 16 GB RAM
```

Verify Ollama is running:

```bash theme={null}
ollama list
curl http://localhost:11434/api/tags
```

## Connect MIRA to Ollama

<Steps>
  <Step title="Start Ollama">
    Ollama starts automatically at login after installation, or run `ollama serve` manually.
  </Step>

  <Step title="Select Ollama in the Engine tab">
    Press ⌘, → **Engine** tab → in the **Provider & Model** section, click the **Ollama (local)** button.
  </Step>

  <Step title="Enter a model name">
    Type the name of any model you have pulled with `ollama pull` into the **Model ID** field.
    MIRA shows a few suggestions as chips, but **any model from `ollama list` works** — just type the exact name.

    | Suggested model | RAM needed |
    | --------------- | ---------- |
    | `llama3.2`      | 8 GB       |
    | `llama3.1`      | 8 GB       |
    | `mistral`       | 8 GB       |
    | `qwen2.5-coder` | 8 GB       |
    | `deepseek-r1`   | 8 GB       |
  </Step>

  <Step title="Set the Base URL (optional)">
    The **Base URL** field appears automatically for Ollama. Default: `http://localhost:11434`. Change it only if Ollama is running on a different host or port.
  </Step>

  <Step title="Save">
    Click **Save & restart bridge**. No API key is needed.
  </Step>
</Steps>

## Performance tips

* Use a Mac with Apple Silicon (M1/M2/M3/M4) — Ollama uses Metal GPU acceleration
* Reduce **Context Budget** in the Engine → NAE settings when using smaller models (e.g. 20 000–40 000 for 7B models)
* Keep other heavy applications closed when running 13B+ models

## Troubleshooting

| Error                          | Fix                                                                      |
| ------------------------------ | ------------------------------------------------------------------------ |
| `ECONNREFUSED localhost:11434` | Ollama is not running — run `ollama serve`                               |
| Model returns garbled output   | Pull the model again: `ollama pull <name>` to ensure a complete download |
| Out of memory                  | Use a smaller model or increase system swap                              |

<Note>
  Edit this page — [Open a pull
  request](https://github.com/satyendra2013/mira-app/edit/main/docs/configuration/providers/ollama.mdx)
</Note>