ANDARTIS

The Nervous System for Personal Intelligence.

📖 Read the Official Documentation

ANDARTIS is a "Local-First" Agnostic Intelligence Framework and Orchestrator. It represents the evolution of ARKHEIN, moving away from slow, general-purpose models towards a distributed ecosystem of hyper-specialized Micro-Models and Conversational SLMs.

The Rebel Vision

The current AI landscape is obsessed with "one model to rule them all." ANDARTIS rejects this. Instead of a massive, slow, and privacy-leaking generalist, ANDARTIS empowers you to build a private fleet of tiny, specialized neural networks trained exclusively on your own data.

• Speed as a Human Right: Micro-models and SLMs (<8B parameters) run on the Apple Neural Engine (ANE) in milliseconds.
• Total Data Sovereignty: No APIs, no telemetry, no clouds. Your machine is the forge; your data is the fuel.
• Hybrid Intelligence: Combines the fluidity of a Senior Analyst (SLM) with the 100% precision of Symbolic Micro-Models.

🚀 Alpha Use Cases: The Revolution of Private Intelligence

1. The Private Detective (Investigation)

Connect chaotic folders of evidence—handwritten notes, discovery docs, and photos.

• Intelligence: Automatically infer timelines from folder dates and map suspect connections across disparate files.

2. The Sovereign Clinician (Medical & Therapy)

Perform high-fidelity patient outcome statistics and clinical research.

• Intelligence: Extract precise dosages from tables and calculate frequency maps of diagnoses with zero hallucinations.

Core Architecture

ANDARTIS is built on a hybrid architecture of a PHP Orchestrator and an Apple MLX acceleration engine, communicating via a zero-latency JSON-RPC STDIO pipeline.

System Components

1. The Orchestrator (Laravel 12 & NativePHP): Manages Intelligence Nodes, configures extraction schemas, and enforces sequential training guardrails to prevent Apple Neural Engine resource contention.
2. The Librarian (Ingestion Pipeline): Crawls folders, hashes files for idempotency, and utilizes the Extraction Forge (Semantic Resonance) to normalize metadata directly into the node's isolated core.sqlite file.
3. The Brains (Capability Plugins): Persistent Apple MLX workers loaded in memory (VRAM) providing domain logic:
   • Senior Analyst (Conversational SLM): Router and intent parser (Intent Blade) that synthesizes raw JSON into conversational prose.
   • Semantic Navigator: Neural search using Siamese Transformers with metadata pre-filtering.
   • Entity Analytic: High-precision symbolic processing and cross-document data aggregation.

The Data Lifecycle

1. Node Initialization: Define an Intelligence Node pointing to a local folder and configure its extraction schema. ANDARTIS creates a dedicated, isolated SQLite database (core.sqlite) exclusively for the node.
2. Ingestion & Extraction: The Librarian ingests files and writes normalized metadata directly to the node's isolated DB.
3. Capability Training: Attachment and sequential training of specialized MLX Brains.
4. Grounded Synthesis: The user queries the node. The Intent Blade (SLM) selects the appropriate tool, the Orchestrator fetches the raw JSON data from SQLite, and the Senior Analyst (SLM) synthesizes it into a zero-hallucination natural response.

---

Tech Stack

• Orchestrator: Laravel 12 & NativePHP (Desktop Application Shell).
• UI: Vue 3 & Inertia.js (Modern Vanilla CSS & Shadcn aesthetics).
• AI Core: Python 3 & Apple MLX (Optimized for Apple Silicon).
• Communication: Persistent JSON-RPC over STDIO (No web servers, zero latency).
• Storage: Isolated per-node SQLite databases (Metadata) + NPZ/Safetensors (Neural Weights).

---

🛠 Getting Started

To run ANDARTIS locally, follow the quick steps below. For a comprehensive, step-by-step setup walkthrough, please consult our official documentation:

👉 Official Getting Started & Setup Guide

Quick Reference

1. Install Host Dependencies:

   composer install
   npm install

2. Configure Python Environment: Create an isolated environment and install the MLX/inference libraries:

   python3 -m venv .venv
   source .venv/bin/activate
   pip install -r python/requirements.txt

3. Fetch Global SLM Weights: Download the quantized senior analyst model weights (~4.5GB):

   python3 python/slm_forge.py

4. Initialize & Launch: Initialize SQLite databases and launch the NativePHP application shell:

   php artisan native:migrate:fresh --seed
   npm run build  # or npm run dev for hot-reloading
   php artisan native:serve

For detailed architecture analysis and our Bring Your Own Dataset (BYOD) model, visit Training vs Node Data.

License

ANDARTIS is open-source software licensed under the MIT license.

Built with ⚡️ by Luca Visciola.