Local AI Models & Hardware

Air-Gapped Hardware Engine

Run specialized literary AI models directly on your hardware. No tokens, zero telemetry, 100% private.

On-Premise Air-Gapped

System RAM

64.0 GB

56.0 GB available

Processor Cores

24 Threads

x86_64 (Microsoft Windows 11 Pro)

Acceleration Engine

NVIDIA GeForce RTX 5080 (16.0 GB VRAM)

NVIDIA CUDA Enabled

Active Model in Litprose AI

Active Google Gemma 2 9B Instruct (Q4_K_M)

Calibrated • Air-Gapped

Downloadable GGUF Language Models (Hugging Face Hub)

Single-file GGUF quantized models optimized for deterministic multi-agent literary prose & translation.

Air-Gap Guard Active • Multi-Process Asynchronous Background Ingestion

Recommended for Literary Prose 4.68 GB

Qwen 2.5 7B Instruct (Q4_K_M)

RAM: 12GB - 16GB

Premier open-weight model calibrated for high-fidelity literary prose, nuanced dialogue, and full-spectrum translation across all 8 supported publishing languages.

Languages: EN, ES, FR, DE, IT, PT, ZH, JA

English Spanish French German Italian Portuguese Chinese Japanese Multilingual (29+)

Deep Stylistic Tone & Voice 5.4 GB

Google Gemma 2 9B Instruct (Q4_K_M)

RAM: 16GB+

Google DeepMind architecture renowned for prose cadence, pacing, and emotional depth in English, Romance, and Germanic languages.

Languages: EN, ES, FR, DE, IT, PT

English Spanish French German Italian Portuguese Multilingual

Model Installed (5494.2 MB)

World Knowledge & Non-Fiction 4.92 GB

Meta Llama 3.1 8B Instruct (Q4_K_M)

RAM: 16GB

Meta flagship open-weight engine with official 8-language multilingual tokenizer, ideal for non-fiction fact checking and complex syntax.

Languages: EN, ES, FR, DE, IT, PT

English Spanish French German Italian Portuguese Multilingual

Ultra-Fast Laptop Friendly 2.1 GB

Qwen 2.5 3B Instruct (Q4_K_M)

RAM: 8GB RAM (Laptops/MacBook Air)

Lightweight model designed for rapid on-premise execution with full 8-language vocabulary and low memory footprint.

Languages: EN, ES, FR, DE, IT, PT, ZH, JA

English Spanish French German Italian Portuguese Chinese Japanese Multilingual (29+)

Model Installed (2007.4 MB)

Creative Writing Engine 1.88 GB

CreativeWriter Llama 3.2 3B (Q4_K_M)

RAM: 8GB - 16GB

Fine-tuned language model acquired from Hugging Face (theprint/CreativeWriter-Llama3.2-3B-GGUF).

Languages: EN, Multilingual

English Multilingual

Model Installed (1925.8 MB)

100% Air-Gapped, Zero-Egress Literary Architecture

Private • Unlimited Ingestion

All language models run entirely on your local machine using GGUF quantization. Your manuscripts, chapters, and author notes never leave your computer—no subscriptions, no API tokens, and zero external telemetry.

GGUF Quantization & RAM Calibration

What is GGUF? A binary model container optimized for fast CPU/GPU inference. It compresses 16-bit neural weights into compact 4-bit or 8-bit representations, allowing large models to fit inside consumer RAM and VRAM.
RAM Safety Envelope: Your system has 64.0 GB RAM and NVIDIA GeForce RTX 5080 (16.0 GB VRAM). Models with Q4_K_M or Q5_K_M quantization run at maximum speed with zero out-of-memory slowdowns.
Model Sizes & Footprints:

• 3B – 4B Models (2 – 3 GB): Ultra-fast drafting, children's dialogue, sentence rephrasing (15–30 tokens/sec).

• 7B – 8B Models (4 – 6 GB): Standard for novels, rich literary fiction, and bilingual translation.

• 12B – 14B Models (7 – 9 GB): High-precision CMOS auditing, complex plot continuity, and historical vocabulary.

Multi-Agent Triad Coordination

When you trigger the Litprose Triad Engine, your active model performs 4 distinct autonomous passes:

1. Creator Agent: Synthesizes prose variations, improves narrative pacing, and expands dialogue.
2. Verifier Agent: Audits against 8 CMOS linguistic categories and book dictionary rules.
3. Supervisor Agent: Calculates the JEV consensus score and heals prose conflicts.

Hugging Face Search & One-Click Ingestion

Search Models (Hugging Face): Click Search Models (Hugging Face) to query thousands of community GGUF models directly from the Hugging Face Hub (TheBloke, bartowski, Qwen, Mistral, Llama 3, Gemma, etc.).
Asynchronous Background Ingestion: Downloads run in a background worker with resume support. You can navigate away, edit books, or generate illustrations while the model downloads seamlessly.
Instant Model Switching: Once downloaded, click Set as Active Engine on any card to immediately switch your active engine across the entire book studio.

Multilingual & Bilingual Fiction Support

Bilingual Mirror Consistency: Models tagged with EN, ES, FR, DE, IT, PT excel at maintaining parallel cadence between English and Spanish spreads without loss of poetic rhythm.
Personal Dictionary Memory: Character names, fantasy places, and unique idioms saved to your personal book dictionary are automatically injected as hard constraints during generation.

Preset Discovery Filters for Authors:

Match My Hardware Profile (64.0 GB RAM)

Host Profile: 64.0 GB RAM • 24 Threads

Select a preset filter above or enter search keywords to discover compatible GGUF models on Hugging Face.