# MegaNova AI

> MegaNova AI is an AI inference cloud powering two things: Libra OS — enterprise digital employees for business operations that run inside your own environment — and Character Studio — AI roleplay characters for creative storytelling. One OpenAI-compatible API gives access to 100+ frontier and open-source models.

## Products
- [Libra OS (enterprise digital employees)](https://meganova.ai/enterprise): On-premise AI platform for business operations — legal, marketing, finance, and operations. Delegate tasks and get source-cited results. Multi-agent orchestration (Libra Kernel), 3-tier AI Firewall, Knowledge DB, 100+ models, SOC2/HIPAA support, ~2-week deployment.
- [MegaNova Cloud (inference API)](https://console.meganova.ai): Frontier-model inference behind one OpenAI-compatible endpoint. Serverless, pay per token, free credits to start.
- [Character Studio (AI roleplay characters)](https://studio.meganova.ai/): Create AI roleplay characters for creative storytelling and character chat — no install, no infrastructure. Powered by the same Libra engine and includes dedicated roleplay models.

## Key pages
- [Home](https://meganova.ai/): Overview of MegaNova AI.
- [Libra OS / Enterprise](https://meganova.ai/enterprise): Digital employees for business operations.
- [Explore models](https://meganova.ai/explore): 100+ models for text, image, audio, video, and roleplay/character chat.
- [Pricing](https://meganova.ai/pricing): Plans and per-token pricing.
- [FAQ](https://meganova.ai/faq): Frequently asked questions.
- [Compute](https://meganova.ai/compute): GPU compute and inference infrastructure.
- [About](https://meganova.ai/about-us): About MegaNova AI.
- [Contact](https://meganova.ai/contact-us): Contact and enterprise inquiries.

## Key facts
- Libra OS is free for personal use; commercial licensing: contact@meganova.ai
- 100+ models behind one OpenAI-compatible API, with intelligent failover and zero vendor lock-in
- On-premise / air-gapped deployment available; data never leaves your network
- 3-tier AI Firewall — regex, ML risk scoring, and an LLM judge — for prompt injection, jailbreaks, and PII
- Character Studio offers dedicated AI roleplay models for creative storytelling

## Benchmarks (measured, ×3-certified)
- System lift: +18.9 points on GAIA (public agentic benchmark, all levels) — the same brain with vs. without Libra OS; +17.2 apples-to-apples on the shipped local brain.
- Grounded scale: 256 grounded Digital Employees per 8×H100 node, at 98.3% grounded success across three runs (measured as grounded success, not HTTP 200).
- Security delta: −35% attack success rate on AgentDojo (26.8% → 17.3% with defenses on), while benign utility holds at 90.7%.
- Method: controlled same-model A/B runs; every number is ×3-certified and reproducible, with full methodology available on request.