“AI wins your trust when it’s guided by your principles”
We are a research-driven team building core foundational technology to unlock trustworthy AI via open efficient language models and multilingual agents designed for safety and reliability. Our technology turns AI from a risk into a company's most reliable employee.
Open multilingual language models
We develop blazing-fast multilingual language models optimised for safety, compliance, and governance.
- Models
- 3
- Languages
- 5+
- Latency
- <100ms
Data sovereignty, open architectures, on-premise deployments
Native coverage with culturally-aware evaluation suites
Optimized for commodity hardware footprints
Our Focus Areas
Why teams partner with us
AI governance can't be an afterthought
Most AI deployments fail compliance audits or drift past safety thresholds undetected. Our principled AI technology goes beyond safety-focused guardrails and acts as a multilingual control layer that governs AI-based solutions in real-time, not only ensuring safety and compliance but also reinforcing brand trust and pursuing business objectives.
Principled Agents
No need to access your AI systems directly: our Principled Agents sit alongside, embedding your company principles to ensure safety, compliance and on-brand decision making in real-time.
Guard Agents
Filter and sanitize risky or sensitive inputs before they reach your AI systems.
Supervisor Agents
Check responses against policy, attach rationales, and escalate exceptions to humans.
Adversarial Agents
Continuously probe deployments with evolving scenarios to monitor and expose regressions early.
Monitor Agents
Analyze traffic and conversations to detect success and failure modes across users, scenarios, and languages.
Meet our founders
Our founders bring a wealth of experience from academia and industry, united by a passion for building AI systems that are not only intelligent but also principled and aligned with human values.
Led knowledge-enhanced LLM research for Siri Information Intelligence at Apple. Former Assistant Professor of Computer Science at Sapienza University of Rome with multiple awards at premier AI/NLP conferences. Co-led pre-training and evaluation of Minerva, Italy's first large language model.Building multilingual AI systems to amplify human potential
Second-time founder and former Assistant Professor of Computer Science Engineering at Sapienza University of Rome. Extensively published in top AI conferences and built cutting-edge AI-powered Information Extraction systems adopted by Big Tech. Co-led pre-training and post-training of Minerva, Italy's first large language model.Advancing AI alignment in real-world applications
Let's get in touch
Whether you need to evaluate model behavior before deployment, implement safety guardrails, or ensure your AI system meets regulatory standards—we're here to help.
Robustness evaluation, probing, and red-teaming
Guardrails, monitoring, and incident response
Mission-critical Generative AI deployments
Language model post-training and alignment
Connect with the founding team
Email us anytime! We triage every message for fit and respond as soon as possible.