First to bring AI Agent + Vision Foundation Model to industrial inspection. Full-stack upgrade — from labeling to inference, with production-grade stability.

Aqrose AIDI Vision Platform 3.4 is officially here.This release is built around two core pillars: efficiency and intelligence. We're proud to introduce — for the first time — AI Agent and Vision Language Model (VLM) capabilities into the platform. The result? Faster inspection throughput, and a major step forward in making industrial vision truly intelligent.

Highlight 1: Labeling just got smarter. Data preparation — significantly faster.

Labeling is often the most time-consuming part of any industrial AI vision project.

AIDI 3.4's smart labeling is completely re-engineered from algorithms to UX — tighter bounding boxes, smoother interactions, fewer misclicks, and a major efficiency boost for both segmentation and detection.

Faster data preparation and a shorter cycle from project initiation to implementation.

Highlight 2: AI Agent is now onboard — customize capabilities just by describing what you need.

AIDI 3.4 introduces the AI Agent assistant for the first time. Just describe your needs in natural language — the AI Agent understands intent and generates the corresponding logic automatically. For simpler scenarios, zero code is all it takes. This is the first step toward a smarter AIDI, with more AI capabilities on the horizon.

Lower barriers to entry, smarter human-machine collaboration.

Highlight 3: Vision Language Model (VLM) integrated — one more powerful tool for tackling complex defects.

AIDI 3.4 brings Vision Language Model (VLM) capabilities into the fold.

With the stronger generalization capabilities of foundation models, AIDI now has a more powerful solution for rapid model switching across multiple product variants. This complements existing small-model capabilities and pushes the boundaries of what's solvable in challenging scenarios.

VLM capabilities now extended to the edge — giving you more confidence in tackling the toughest inspection scenarios.

Highlight 4: Defect-free learning — lock in anomalies with OK samples only.

EL anomaly detection eliminates the need for large defect sample sets. With just a small number of OK images, you can complete model training and quickly determine whether a product has obvious defects — with OK/NG results automatically output.

This module is tailored for scenarios such as material mix-ups and significant structural defects — balancing training and inference efficiency. It provides a lightweight, cost-effective vision inspection solution for rapid production line screening, helping businesses quickly establish anomaly detection capabilities.

Just a handful of good samples are enough to quickly build a detection model — drastically lowering the barrier to entry and shortening project deployment timelines.

No need to wait for defects to appear. Just a few good samples — and you're ready to start screening.

And that's not all. AIDI 3.4 also brings a ~5x boost in training efficiency, auto-split view for segmentation tools, plus multiple performance and stability enhancements for large-scale projects.

From AI Agent to Vision Language Model, AIDI 3.4 builds on efficiency and stability — and officially takes that critical step toward true intelligence.

Request a trial now.