Enabling Agentic AI through
Perception, Reasoning & Action

Phronetic’s agentic platform is built on the backbone of continuous, applied research. We develop and open-source small, powerful models that fuel the next generation of multimodal agents — enabling them to see, understand, and act in complex real-world environments.

Owlet

Lightweight Vision-Language Models for Real-Time Video Understanding

Owlet is a family of fine-tuned vision-language models (VLMs) optimized for real-time video analytics. These models are specifically designed for use cases such as:

These models are specifically designed for use cases such as:

  • Safety event detection (e.g., fire, fall, assault)

  • Human activity recognition in CCTV feeds

  • Visual Question Answering (Video-QA)

  • Multilabel video classification

Owlet variants are lightweight enough for edge deployment yet powerful enough to provide contextual video insights at scale. They're integrated across products like ABM and ClipGen and are foundational to our field agent capabilities.

View on Hugging Face

RZN

Reasoning Engine for Task Planning and Agentic Intelligence

RZN is Phronetic's collection of reasoning-first models built to power intelligent task execution across our agents. These models enable AI agents to break down complex instructions, use tools and memory effectively, and adapt in real time to dynamic tasks.

The RZN models support:

  • Multi-step reasoning and goal planning

  • Tool calling and integration chaining

  • Agentic workflows with memory and retry logic

  • Decision-making across HR, Finance, Research, and more

RZN is foundational to our Desk Agents and developer platform, helping agents not just respond — but reason, plan, and act.

View on Hugging Face

Theia

Video-First Multimodal LLM for Visual Scene Understanding

Theia is a proprietary large language model that processes and interprets visual inputs like video, screen shares, and image sequences.

It powers our flagship ABM and developer tools platform, enabling capabilities such as:

  • Deepfake detection and content integrity checks

  • Screen-share analysis and chat-based video guidance

  • Visual question-answering and summarization

  • Real-time multimodal inference for agents

Theia bridges the gap between LLMs and video-first data, making it ideal for applications in surveillance, enterprise ops, and virtual collaboration

Research Blogs

Owlet-HAR-1: Building Better VLM for Human Activity Recognition

Introducing Owlet-HAR-1, a specialized Vision Language Model fine-tuned for human activity recognition in videos.
Read more

Owlet-Safety: Lightweight Model for Video Safety Monitoring

A lightweight fine-tuned vision-language model for detecting fire, assault, theft, and other safety threats in video, open-source, efficient, and ready for real-world deployment.
Read more

Scaling Reasoning in VLMs with Reinforcement Learning

A practical exploration of how reinforcement learning can enhance reasoning in multimodal models - demystifying DAPO, GRPO and their real world impact
Read more

10 Must-Have Free AI Tools For Video Editors To Skyrocket Your Social Media Engagement

In today world, there’s no time for mindless searching or waiting. So here are 10 video tools that you can use to create videos easily & quickly.
Read more

AI-Powered Solutions for Law Enforcement: Enhancing Traffic Violation Detection

Traffic management systems can now be transformed with AI. Read this blog by Phronetic AI to learn about AI-powered traffic enforcement.
Read more

Owlet-Phi-2-Audio: Our latest multimodal AI model to jointly understand audio and visual signals

We propose a model architecture that handles audio-visual inputs explicitly. We train our model with both audio and visual data from a video instruction-tuning dataset. Comparison with vision-only baselines, and other audio-visual models sh
Read more
View all