Lustrew Dynamics builds AEGIS, an embodied AI platform that combines vision, OCR, and tool-use capabilities to automate complex screen‑based workflows. Their first product, ADA, offers a sandboxed AI operator for desktop and web environments, enabling enterprises to execute multi‑step tasks, improve accessibility, and accelerate decision‑making without custom scripting.
Funding
Funding not disclosed
Founders
Product
Problem
Current AI solutions are fragmented, handling only single modalities such as text or images, and lack the ability to perceive, reason, and act across diverse digital environments. This limits automation of complex, screen‑based workflows and reduces the efficiency of enterprises that need integrated, end‑to‑end AI operators.
Solution
Lustrew Dynamics is developing AEGIS, an embodied general‑intelligence architecture that unifies high‑accuracy vision, OCR, peripheral screen awareness, and tool‑use capabilities into a single coherent agent. The system combines deep‑learning perception models (e.g., DeepSeek, InternVL) with a Qwen‑based reasoning engine and modular tool interfaces for web search, file manipulation, and UI interaction. By maintaining contextual memory and self‑consistent reasoning loops, AEGIS can understand documents, navigate software interfaces, and execute multi‑step workflows without predefined templates. The first public product, ADA, will expose these capabilities as a sandboxed AI operator for desktop and web environments, enabling enterprises to automate complex tasks, improve accessibility, and accelerate decision‑making.
Target Audience
Primary customers are enterprise teams seeking to automate screen‑based processes—such as finance, healthcare, compliance, and cybersecurity—as well as government and research organizations that require an AI operator capable of integrated perception and action.
Features
- Integrated vision pipeline delivering near‑perfect OCR with layout preservation and real‑time text playback
- Peripheral vision module that identifies UI elements across full‑screen screenshots with ~90% accuracy
- Tool‑use framework supporting live web searches, local file creation/editing, screenshot capture, and structured workflow execution
- Contextual memory and reasoning loops that maintain task continuity and intent across multimodal inputs
- Secure, sandboxed execution environment for desktop and web access, ensuring safe operation of autonomous actions
- Extensible modular architecture allowing addition of new capabilities and domain‑specific tools