aurarouter 0.1.0

Multi-model MCP routing fabric for local and cloud LLMs

AuraRouter: The AuraXLM-Lite Compute Fabric

Current Status: Production Prototype v3 (Feb 2026)
Maintainer: Steven Siebert / AuraCore Dynamics

Overview

AuraRouter implements a simplified role-based configurable xLM (SLM/TLM/LLM) prompt routing as an MCP server. AuraRouter is designed to orchestrate local and cloud resources for AuraCore development. It acts as an intelligent middleware for an MCP Client (ie Gemini CLI), allowing you to route code generation tasks to local hardware while maintaining a cloud safety net.

It implements an Intent -> Plan -> Execute loop:

1. Router: A fast local model classifies the task (Simple vs. Complex).
2. Architect: If complex, a reasoning model generates a sequential execution plan.
3. Worker: A coding model executes the plan step-by-step.

Architecture

graph TD
    User[Gemini CLI] -->|Task| Router{Intent Analysis}
    Router -->|Simple| Worker[Coding Node]
    Router -->|Complex| Architect[Reasoning Node]
    Architect -->|Plan JSON| Worker
    
    subgraph Compute Fabric [auraconfig.yaml]
        Worker -->|Try| Node1[Local RTX 3070]
        Node1 -->|Fail| Node2[Cloud Fallback]
    end

Prerequisites

• Python 3.12+
• Ollama (Running locally with qwen2.5-coder:7b or similar)
• Google AI Studio Key (For cloud fallback/reasoning)

Installation

1. Environment Setup

Create the isolated environment with the required dependencies.

conda env create -f environment.yaml
conda activate aurarouter

2. Pull Local Models

We recommend the Qwen 2.5 series for consumer hardware speed/stability.

ollama pull qwen2.5-coder:7b

3. Configuration

Edit auraconfig.yaml to define your nodes and paste your API keys.

models:
  local_3070:
    provider: ollama
    endpoint: http://localhost:11434/api/generate
    model_name: qwen2.5-coder:7b

  cloud_gemini:
    provider: google
    model_name: gemini-2.0-flash
    api_key: "AIzaSy..." # Paste key here

4. Auto-Registration

Run the installer to inject AuraRouter into your Gemini CLI settings.

python aurarouter.py --install

Follow the prompts to auto-detect your settings.json.

Usage

Restart your Gemini CLI. You can now use natural language to trigger the fabric.

The "Fast Lane" (Local Only):

"Write a python function to calculate Fibonacci." (Routes directly to Local Qwen)

The "Heavy Lane" (Cloud Plan + Local Build):

"Create a distributed lock manager in C# with an interface and unit tests." (Routes to Cloud Architect for planning, then Local Qwen for execution)

Scaling Guide

When you add new on-prem xLM resources:

1. Open auraconfig.yaml.
2. Uncomment the local_3090_deepseek block under models.
3. Add it to the top of the reasoning role list.
4. Restart the router. No code changes required.

Troubleshooting

• "Empty response received": The local model is likely OOMing or timing out. Check the timeout setting in auraconfig.yaml.
• "Model not found": Ensure the model_name in YAML matches ollama list exactly.
• Installer fails: Manually add the mcpServers block to your ~/.geminichat/settings.json pointing to the aurarouter.py absolute path.

License

Copyright 2026 AuraCore Dynamics Inc.