Give every enterprise
its own dedicated agent
Twenty years of AI work across China, the US and Europe — and our enterprise-grade agent foundation, CATA OS.
CATA Assistant
Online · answers from this site
The people behind the platform
Twenty years of AI research and delivery across China, the US and Europe. The platform is what they learned shipping it.

Shu Chen
Co-founder & Chief Scientist
Focuses on ML systems and large-scale distributed inference; leads the agent orchestration kernel of CATA OS.

Jing Lin
Partner, Enterprise Transformation
Brings consulting methodology into AI delivery; long experience in organizational change for manufacturing and retail.

Yiming Zhao
Head of Research
Works on agents for scientific discovery, co-building research workflows with university labs.

Jingjing Liu
University Professor · PhD in Computer Communications
PhD in computer communications and a university professor. Long-standing research in machine learning, communication algorithms and embedded systems; author of books on IoT embedded technology.
Research we run with university labs
Two tracks: agents that understand an industrial shop floor, and agents that speak the language of control logic.
Tsinghua University
Industrial agents
Agents that read the shop floor
With the lab, we research agents that read production data, process knowledge and equipment manuals, and turn them into judgements on scheduling, quality inspection and maintenance.
- Industrial knowledge bases and document understanding
- Task orchestration across agents
- Line-level anomaly diagnosis and recommended actions
Norwegian university
PLI agents
Control logic, spoken to a model
With the professor's team, we research PLI-related agents that combine control logic with language models, aimed at shipbuilding and other industrial-control scenarios.
- Semantic modelling of PLI signals
- Control logic working with language models
- Safety boundaries for industrial control
A bespoke AI OS — your decision brain, and the workbench you run it on
The CATA enterprise workbench isn't another back office. It turns facts scattered across stores, systems and spreadsheets into a decision brain, then turns those decisions into actions people take every day.
XX Department Store · CATA Workbench
Sample interfaceEast China · Today
18 stores · 11 Nov, 10:24
Sales today
¥3.84M
+8.4%vs last week
Footfall
42,180
+3.1%vs last week
Conversion
28.6%
−1.2ptvs last week
Stores to fix
6
+2since yesterday
Sales by hour · today
Peak 18:00 · ¥482k
Needs your attention
View all 6Nanjing XinjiekouHigh
Conversion below regional average for 3 days
Hangzhou HubinMed
Womenswear inventory turns in 48 days
Suzhou GuanqianMed
2 return complaints today
- Decision brain
- Facts from stores, systems and spreadsheets, reconciled into one set of metrics.
- Flagged tasks
- Anomalies become the actions someone should take today — not another report.
- Workbench agent
- It knows this store's numbers, the last adjustment and why it was made.
AI hardware
Catalyst runs partner labs in Norway, the Netherlands and Germany, focused on applied AI hardware: AI hearing earbuds, a desktop pick-and-place machine, a full-colour 3D printer and more.
Desktop SMT Pick-and-Place
0402 placement accuracy, on a desk.
A desktop pick-and-place machine for prototyping and small-batch builds. Vision-guided 0402 placement has been demonstrated on a test setup; the complete machine is still in development, and 1000 CPH / ±50 µm are whole-machine targets rather than measured results.
Product detailsAI Hearing Earbuds
Earbuds that make speech clear — and keep a voice assistant close at hand.
TWS earbuds combining on-device speech enhancement with a built-in voice assistant. Concept and industrial-design mock-up are complete; a prototype is the next milestone. Design targets: 8 hours per charge plus 8 more from the case, at roughly USD 20 material cost per unit.
Product detailsWhat we shipped and learned lately
Product releases, academic collaboration, and what we learn putting agents into production.

12/10/2026Product
CATA OS Ships Orchestration Kernel 2.0
Adds multi-agent collaboration and unified evaluation, cutting time-to-production from months to weeks.

18/09/2026Academic
Joint Research with a Norwegian University Lab
A two-year collaboration on evaluation methods for agents in scientific discovery.
What changed in production

Research
Research Agent Workflow for a University Lab
Rebuilding literature review and experiment logging with agents, cutting repetitive work to a third.
65%
Less repetitive work

E-commerce
Multi-Agent Operations Hub for a Maternal & Infant Brand
Assortment, content and service run through coordinated agents; content output tripled.
3×
Content output

F&B Chain
Store Performance Analytics for a Restaurant Chain
Turning store data into daily actionable moves; per-store labor efficiency up 40%.
40%
Labor efficiency