> ## 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.

# RLM Settings

> Configure the Recursive Loop Machine — provider, model, iteration limits, and concurrency.

The Recursive Loop Machine (RLM) settings control the LLM provider, model, generation parameters, and concurrency behaviour.

## Opening RLM settings

Press ⌘, → **Engine** tab. If RLM is your active engine (or you want to configure it), its settings appear in the lower panel.

## Provider & Model

The RLM supports four providers: **AWS Bedrock** (default), **Anthropic**, **OpenAI**, and **Ollama**. Select a provider with the four provider buttons in the **Provider & Model** section, then enter the **Model ID**.

**You can use any model from a supported provider** — MIRA shows suggestion chips as a starting point, but the Model ID field accepts any valid model ID your account can access.

<Warning>
  The RLM Engine is designed and tested to work reliably with **Claude** and **GPT** series models.
  Other models may produce inconsistent results or fail to follow the reasoning loop correctly.
</Warning>

| Provider    | Default credential                   |
| ----------- | ------------------------------------ |
| AWS Bedrock | AWS credentials (Bedrock / AWS tab)  |
| Anthropic   | `ANTHROPIC_API_KEY` (API Tokens tab) |
| OpenAI      | `OPENAI_API_KEY` (API Tokens tab)    |
| Ollama      | None (local)                         |

For Ollama, a **Base URL** field appears (default: `http://localhost:11434`). For Bedrock, an **AWS Region** dropdown appears.

## Generation parameters

| Setting               | Range      | Default | Description                                            |
| --------------------- | ---------- | ------- | ------------------------------------------------------ |
| **Max iterations**    | 1–50       | 20      | Hard cap on reasoning loop iterations per query.       |
| **Temperature**       | 0–1        | 1.0     | 0 = deterministic · 1 = default model creativity.      |
| **Max output tokens** | 512–32 768 | 8 096   | Maximum tokens per LLM response in the reasoning loop. |

## Concurrency & Behaviour

| Setting                                 | Default | Description                                                                 |
| --------------------------------------- | ------- | --------------------------------------------------------------------------- |
| **Max concurrent sessions**             | 3       | How many sessions can run queries at the same time (1–5).                   |
| **Background completion notifications** | On      | Show a toast when a session finishes while you are viewing a different one. |
| **Auto-focus on completion**            | Off     | Automatically switch to the session that just finished.                     |
| **Verbose mode**                        | Off     | Stream all RLM iteration output to the console for debugging.               |

## Runtime

| Setting               | Default   | Description                                                 |
| --------------------- | --------- | ----------------------------------------------------------- |
| **Python executable** | `python3` | Path to the Python 3.11+ binary used to run the RLM bridge. |

MIRA ships with a bundled Python 3.11 runtime that is used automatically. You only need to change this if you are running a custom Python environment.

## Saving changes

Click **Save & restart bridge**. The RLM Python bridge restarts automatically with the new settings.

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