Skip to main content

Introducing ONE WARE Agents: EU Hosted AI Agents That Build Fast, Accurate Vision Models

· 4 min read
Hendrik Mennen
Hendrik Mennen
CTO

ONE WARE Agents build custom ONE AI vision models for you, right inside ONE WARE Studio. Run the agents with your ONE WARE account in EU data centers, bring your GitHub Copilot subscription, or use self-hosted options.

Describe what you want to detect, and ONE WARE Agents handle the rest: from dataset creation to model export for your hardware.

The Agent Builds the Model, Your Hardware Runs It​

ONE WARE Agents don't run inference on your production line. Large AI models are far too slow and expensive for high-speed inspection. Instead, the agent acts as a vision AI engineer: it uses ONE AI to generate a specialized architecture tailored directly to your data and target hardware.

The resulting model:

  • Is faster and more accurate: ONE AI models usually beat universal networks, for example with 10× fewer errors than YOLOv8 or 24× fewer errors on a €45 FPGA. The detector in the demo video runs at >500 fps with 1.8 ms latency on a CPU.
  • Fits your hardware: from microcontrollers and FPGAs to CPUs and GPUs (supported hardware).
  • Runs 100% locally: Zero cloud dependency, zero internet connection required, zero agent overhead.
  • Integrates anywhere: Export directly to ONNX, TensorFlow Lite, or synthesizable VHDL.

What ONE WARE Agents Can Do​

With access to over 40 native tools, ONE WARE Agents don't just chat. They operate ONE WARE Studio and ONE AI directly:

Projects & Settings

Sets up your ONE AI project and finds the right settings for your task and your target hardware.

Create new projectsChoose classification, detection or segmentationSwitch between Basic and Advanced modeManage configurations for different targetsTune model and hardware settings

Labels & Image Processing

Defines what your model should find and how your images are prepared for training.

Add, rename and recolor labelsClean up unused labelsAdd prefiltersAdd augmentationsReorder the processing pipeline

Dataset & Import

Brings your data into the project and makes sure it is ready for training.

Turn videos into training imagesImport YOLO, COCO, Pascal VOC and mask datasetsFind missing labels and split problemsAnalyze object sizes and class balance

Auto-Labeling & Review

Labels your images automatically, so you only have to check and correct the result.

Label datasets automatically with SAMPre-label images with your trained modelOpen the annotation tool for your review

Synthetic Data

Creates training data when you have little or no real images, so you can train a first model right away.

Generate images from a descriptionAdd defects, change backgrounds and lightingBuild fully labeled synthetic datasetsCreate test videos

Training, Test & Export

Trains your model in OneWare Cloud and delivers a tested, ready-to-use model.

Train in OneWare CloudFollow the training progressExplain failed trainingsTest models with metrics and visualizationsExport and download ONNX models

Files & Terminal

Works with files and tools in OneWare Studio like a developer would.

Read and edit project filesWrite and run Python scriptsRun commands in the terminal

Beyond tool access, agents incorporate built-in ONE AI domain rules: recommending resolution based on feature size, enforcing clean validation splits, tuning augmentation pipelines, and analyzing training metrics to suggest improvements.

EU Hosted, Self-Hosted, or Your Own Subscription​

Choose the provider and hosting model that fits your security requirements:

  • EU Hosted Models: Processed exclusively in European data centers (GPT-6.1 Sol (EU), GPT-6 Luna (EU), DeepSeek V4 Flash (EU), FLUX.2 Pro (EU)). Organization admins can enforce an EU Only Policy under Settings → Policies to block any data from leaving the EU.
  • Self-Hosted Models: Run the agent on your own model server, for example with Ollama or any OpenAI-compatible endpoint, so the chat never leaves your network.
  • GitHub Copilot: Sign in with your existing GitHub account and use your Copilot plan directly inside ONE WARE Studio.
  • Bring Your Own Key (BYOK): Connect OpenAI, Anthropic, or any OpenAI-compatible API with your own API keys.

ONE WARE Agents models and prices in the organization policies

Full Data Privacy, End to End​

Combine a self-hosted agent with a self-hosted training worker, and your images, prompts and models never leave your infrastructure. The finished model runs offline on your device, without the agent and without an internet connection.

Synthetic Data with Image Models​

Need a model before having hardware or physical samples? ONE WARE Agents combine language models with generative image models like FLUX.2 Pro (EU) and GPT Image 2.5 Flare:

  • Generate training samples from text prompts or reference photos.
  • Simulate defects, lighting shifts, and background variations.
  • Auto-label synthetic assets automatically for immediate training.

Get Started​

ONE WARE Agents are available in all Pro plans and the Pro Trial, as well as with existing GitHub Copilot or BYOK setups:

  1. Update the ONE AI extension in ONE WARE Studio.
  2. Open View → Tool Windows → AI Chat.
  3. Select your provider (ONE WARE Agents, GitHub Copilot, or BYOK).
  4. Switch to Agent mode and describe your task.

Need a trial or have questions? Book a demo or reach out at sales@one-ware.com. For technical setup, visit the ONE WARE Agents documentation.