AuraRouter: The AuraXLM-Lite Compute Fabric

Current Status: Production Prototype v5 (Feb 2026) Maintainer: Steven Siebert / AuraCore Dynamics

Overview

AuraRouter implements a role-based configurable xLM (SLM/TLM/LLM) prompt routing fabric. It acts as intelligent middleware that routes tasks across local and cloud models with automatic fallback. AuraRouter is content-agnostic -- it handles code generation, summarization, analysis, RAG-enabled Q&A, and any other prompt-based work. It can run as an MCP server, a desktop GUI application, or a managed service on AuraGrid.

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

1. Classifier: A fast local model classifies the task (Direct vs. Multi-Step).
2. Planner: If multi-step, a reasoning model generates a sequential execution plan.
3. Worker: An execution model carries out the plan step-by-step.

Architecture

graph TD
    User[MCP Client / GUI] -->|Task| Classifier{Intent Analysis}
    Classifier -->|Direct| Worker[Worker Node]
    Classifier -->|Multi-Step| Planner[Planner Node]
    Planner -->|Plan JSON| Worker

    subgraph Compute Fabric [auraconfig.yaml]
        Worker -->|Try| Node1[Local Model]
        Node1 -->|Fail| Node2[Cloud Fallback]
    end

Installation

1. Core Install (Required)

pip install aurarouter

2. Local Backend (Plugin-based Architecture)

AuraRouter uses a Modular Backend Architecture to keep the core package lean and platform-independent. To use local inference (via llama.cpp), you must install at least one "flavor" package.

AuraRouter dynamically discovers all installed backend plugins at runtime and automatically selects the best one based on your local hardware (preferring GPU over CPU). See BACKEND_PLUGINS.md for technical details.

Platform / Hardware	Recommendation	Command
NVIDIA RTX 50/40/30/20	aurarouter-cuda13	pip install aurarouter-cuda13
Older NVIDIA GPUs	aurarouter-cuda12	pip install aurarouter-cuda12
Windows (CPU Only)	aurarouter-win-x64	pip install aurarouter-win-x64
MacOS (M1/M2/M3/Intel)	aurarouter-macos-x64	pip install aurarouter-macos-x64
Linux / Generic GPU	aurarouter-vulkan	pip install aurarouter-vulkan

You can install multiple backends; AuraRouter will score each one using its internal diagnostics and use the most performant one for your current machine.

Optional Dependencies

# With HuggingFace model downloading
pip install aurarouter[local]

# Everything (Core + Dev tools)
pip install aurarouter[all]

Source Install

git clone https://github.com/auracoredynamics/aurarouter.git
cd aurarouter
pip install -r requirements.txt        # Core dependencies
pip install -r requirements-local.txt   # Optional: local inference deps
pip install -e .                        # Editable install

Conda

conda env create -f environment.yaml
conda activate aurarouter

Standalone Executable (Windows)

If you prefer to run AuraRouter without a Python environment, you can build a standalone executable. You can choose to bundle specific backends or all of them:

# Build with all installed backends
python build.py

# Build a universal executable (bundles all potential backends)
python build.py --backends all

# Build for specific hardware only
python build.py --backends cuda13,win-x64

The executable will be generated at aurarouter/dist/aurarouter.exe.

See DEPLOYMENT.md for detailed deployment and configuration guide.

Quick Start

1. Configuration

Run the interactive installer to create a config template:

aurarouter --install

Or manually create ~/.auracore/aurarouter/auraconfig.yaml:

models:
  local_qwen:
    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..."

roles:
  router:   [local_qwen, cloud_gemini]
  reasoning: [cloud_gemini]
  coding:   [local_qwen, cloud_gemini]

# Optional: semantic verb synonyms for intent classification
semantic_verbs:
  coding:
    synonyms: [programming, code generation, developer]
  reasoning:
    synonyms: [planner, architect, planning]

2. Run

# Standalone executable (Windows)
.\aurarouter\dist\aurarouter.exe        # Run MCP server
.\aurarouter\dist\aurarouter.exe gui    # Launch desktop GUI

# Python module (Installed)
aurarouter                             # Run MCP server
aurarouter gui                         # Launch desktop GUI

Providers

Provider	Type	Config Key	Dependencies
Ollama	Local HTTP	ollama	None (uses httpx)
llama.cpp Server	Local HTTP	llamacpp-server	None (uses httpx)
llama.cpp (Managed)	Local Native	llamacpp	Bundled binary (no extra install)
OpenAPI-Compatible	Local/Cloud HTTP	openapi	None (uses httpx)
Google Gemini	Cloud	google	Included
Anthropic Claude	Cloud	claude	Included

The OpenAPI provider works with any OpenAI-compatible API endpoint (vLLM, text-generation-inference, LocalAI, LM Studio, etc.).

GUI

The desktop GUI (included in the base install) provides:

• Singleton enforcement — Only one AuraRouter instance runs at a time; subsequent launches detect the existing instance
• Environment selector — Switch between Local and AuraGrid deployments at runtime
• Service controls — Start, stop, and pause the MCP server or AuraGrid MAS
• Model loading progress — Visual progress indicator when local GPU models are loading
• Execute tab — Task input with file attachment, DAG execution visualization, task output
• Models tab — Local GGUF model browser, HuggingFace downloads, local file import, grid model listing (AuraGrid)
• Configuration tab — Model CRUD with capability tags, fallback chain editor with known roles, semantic verb configuration, YAML preview, cell-wide save warnings (AuraGrid)
• Grid panels (AuraGrid) — Deployment strategy editor, cell node status, event log
• Health dashboard — Per-model health status with clickable indicator (state-aware: reports correctly when service is stopped)
• Privacy-aware routing — Automatically re-routes prompts containing PII away from cloud models to local/private-tagged models
• Prompt history — Last 20 tasks with results, restorable from dropdown
• Keyboard shortcuts — Ctrl+Enter (execute), Ctrl+N (new), Escape (cancel)

All configuration changes are persisted to auraconfig.yaml. See GUI_GUIDE.md for the complete guide.

CLI Commands

Command	Description
aurarouter	Run MCP server (default)
aurarouter gui	Launch desktop GUI
aurarouter backends	List discovered backends and hardware health
aurarouter --rescan-hardware	Clear backend cache and re-run hardware diagnostics
aurarouter download-model --repo REPO --file FILE	Download GGUF model from HuggingFace
aurarouter list-models	List locally downloaded GGUF models
aurarouter remove-model --file FILE	Remove a downloaded model
aurarouter auto-tune --file FILE	Suggest optimal parameters for a GGUF model
aurarouter --install	Interactive installer for MCP clients
aurarouter --install-gemini	Register for Gemini CLI
aurarouter --install-claude	Register for Claude

MCP Tools

Dynamic Model Registration

External services can register new GGUF models via the aurarouter.assets.register MCP tool:

# Example: Register a fine-tuned model
result = mcp_client.call_tool("register_asset", {
    "model_id": "my-finetuned-qwen",
    "file_path": "/path/to/model.gguf",
    "repo": "myorg/my-finetuned-model",
    "tags": "coding,local,fine-tuned"
})

After registration, add the model to a role chain and restart the MCP server to enable routing.

Asset Discovery

Query available local GGUF files via the aurarouter.assets.list MCP tool:

# List all downloaded models
result = mcp_client.call_tool("list_assets", {})

Returns a JSON array of asset entries with repo, filename, path, size, and metadata.

AuraGrid Integration (Optional)

AuraRouter can be deployed as a Managed Application Service (MAS) on AuraGrid for distributed access to routing services.

pip install aurarouter[auragrid]

See AURAGRID.md for the complete integration guide.

Scaling Guide

When you add new on-prem xLM resources:

1. Open auraconfig.yaml (or use the GUI Configuration tab).
2. Add the new model under models.
3. Add it to the appropriate role chain under roles.
4. Restart the router (or save from the GUI). 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.
• "huggingface-hub is required": Run pip install aurarouter[local] to enable model downloading from HuggingFace.
• "llama-server binary not found": AuraRouter needs a backend plugin to run local models. Install a backend package (e.g., pip install aurarouter-cuda13) or set the AURAROUTER_LLAMACPP_BIN environment variable to point to a custom llama-server binary.
• PySide6 issues on headless servers: PySide6 is a core dependency. On headless/server-only deployments, use the MCP server mode (aurarouter) which does not launch the GUI.

License

Copyright 2026 AuraCore Dynamics Inc.