One general world model
A shared representation of physical reality: space, material, light, motion, causality. Four product lines run on the same foundation, the same self-training platform, the same compute scheduling layer.
OUR THESIS
AGI = one general world model + N vertical small models.
Not "a large language model plus skills or agents." Language models are strong at composing symbols and knowledge. But manufacturing, robotics and surveillance ask a different question: how an object exists in three-dimensional space, how it moves, and why it moves that way — constraints that live in the physical world, not in a text corpus. Bolting tools onto a language model changes what it can call; it does not give it a representation of the physical world.
N vertical small models
One small, fast model per scenario — inference at the edge, retrained from what that site actually sees. The small model owns precision and latency; the foundation keeps them all reading the same physical world.
Why not LLM + skills
Tool-calling answers "can it operate this." A world model answers "can it see what is there." A defect verdict, a grasp pose, a predicted trajectory are not retrieved — they are computed from 3D structure and physical law.
OUR APPROACH
Knowledge lives on the data side, not piled into the weights.
At DaoAI, moving to a new scene does not mean retraining a bigger model. The base model takes a handful of good-part samples and amplifies them, under physical constraints, into enough training data. A 1–20 sample cold start, a model trained in 30 seconds, a vertical derived with zero lines of code — these follow from that approach, not from parameter count.
ONE FOUNDATION · DAOAI WORLD
So the world model has to understand these five together.
A conventional vision system solves one question at a time: is there a defect here, where is this point. DaoAI World holds space, material, light, motion and causality in a single model — which is why four product lines can share one foundation, and why the same system can move from line-side defect detection to quantitative analysis of electron-microscope images.
3D structure
Reconstructs a scene's geometry and scale from 2D pixels rather than template-matching on a plane. Our own 3D camera repeats to 4μm.
Reflective and black parts
DaoAI's 3D structured light plus HDR imaging — black, highly reflective and complex curved surfaces imaged cleanly in one shot, without powder spray or exposure stacking.
Changing conditions
The model recognises the object, not one particular pixel distribution, so shifting plant light or a new material batch doesn't mean starting over.
Pose and trajectory
Hand-eye calibration to ±0.05mm, recognition and locating in under a second, direct bin picking, no point-by-point teaching.
Why it moves that way
Infers how and why objects move by the laws of physics, turning "understanding" into something a robot can actually execute.
All four product lines share one foundation, one self-training platform and one compute-scheduling layer — from inspection and robot vision to SkyVision, and on to the generative side with Wemio.Explore DaoAI World
MANIFESTO · SOVEREIGN PHYSICAL AI
The strongest AI
shouldn't ask you to hand over your data.
Chat, images, code — that's digital AI. Reading a defect, directing a robot, generating a world that obeys physics — that's physical AI. And physical AI lives on the line, inside the plant, where the data shouldn't take a single step outside.
DaoAI's answer is blunt: the whole system can run inside your own server room — data stays in the plant, the model trains itself, and the capability stays best-in-class. 100% on-premise · no public cloud required · flexible SDK / API / Docker integration.
WHY NOW
Keeping data on site shouldn't cost you capability.
For government, transport and energy — critical information infrastructure — the revised Cybersecurity Law's localisation requirement is already a hard constraint. Sovereign AI has to hold three things at once: the data stays in your plant · you own the technology chain · you decide how it deploys. That is why we build the whole stack, from the sensor to the world model.
Data sovereignty
100%
on-premise, no public cloud requiredLine data stays inside your plant, meeting enterprise data-security and compliance needs.
Technology sovereignty
4μm
repeatability, our own 3D camera3D structured light plus HDR imaging — black and reflective parts imaged cleanly in one shot, full stack in-house from sensor to world model.
Operational sovereignty
3
SDK / API / Docker integrationsEmbeds into any equipment and your existing line systems — deployment shape and pace are yours to set.
THE CONSOLE
Three tabs — the three screens your operators live in.
Review components, judge defects, compare against golden. Three real DaoAI World workflows. Click a tab.
OK
OK
Good
Component Review
A thumbnail grid to walk the batch. Selection is green-bordered, verdicts sit on the image — the operator only confirms or overrides.
Batch size1,248
Passed1,241
Awaiting review7
AI verdict
Polarity, offset and missing-part checks all within tolerance.
Bridge
Bridge
Particle
Not good · review
Defect Detection
Surface defects overlay the original image, coloured by class. Threshold sliders are adjustable on the line and take effect immediately — no retraining.
Defects found3
Largest area0.42 mm²
Confidence0.96
AI verdict
Two solder bridges and one foreign particle — rework recommended.
R12 missing
C7 offset
Defective
Golden Sample
Side by side: the current sample against the golden reference, with differences boxed automatically — this is what "model from one good sample" looks like in the UI.
ReferenceGOLDEN-0417
Difference regions2
VerdictDEFECTIVE
AI verdict
R12 missing and C7 offset out of tolerance — defective.
Interface illustration — shown to explain the workflow and verdict logic. Values are demonstration samples, not any single customer's real batch data.
ONE FOUNDATION · FOUR PRODUCT LINES
One world model, four product lines.
Space, material, light, motion, causality — DaoAI World understands them together, which is why reading manufacturing · guiding robots · reading cities · generating content all run on the same foundation.
Manufacturing · Inspection
Industrial ACI (AI-AOI)
Traditional AOI runs on hand-written rules and thousands of defect samples, reprogrammed board by board. We use feature cognition from a visual foundation model: one good sample to start, then line feedback flows back and it keeps sharpening.
No CAD drawings, no pre-collected defect library
Auto-programs and goes live in about 5 minutes — operators run it themselves
PCBA / SMT / semiconductor / display / connector inspection
98%Detection accuracy straight out of auto-programming
5 minOne good sample to live
3Software · 2D · 3D systems
Smart mfg · Robotics
Robot Vision
Eyes that understand space. 3D-vision-guided picking, machine tending and direct bin picking — it reads the scene, plans its own path, and holds sub-millimetre hand-eye coordination.
Stable on reflective metal, black parts and complex curved surfaces
Works with structured-light / ToF / stereo input, no point-by-point teaching
Works with major robot brands — inspection and guidance share one foundation
±0.05mmHand-eye calibration
<1sRecognition & locating
100%On-premise
Cities · Surveillance
SkyVision
A no-code visual AI platform. A few on-site sample images, a few minutes of training, edge inference — the video stream never leaves your network, and it reuses the cameras you already have.
No code: your own staff label, train and deploy
Edge deployment — video never leaves the local network
Works with existing cameras and NVRs, low retrofit cost
256Concurrent streams
<12msEdge inference latency
99%RTSP / ONVIF compatible
Forward-looking · Content
Wemio · enterprise content platform
A content platform for corporate marketing teams — company IP, brand films, founder IP, product explainers and company story dramas, all in one workspace. It is also the same foundation extended toward generative work: inspection is a discriminative vision task and video generation is a generative one — their model objectives and evaluation regimes could hardly differ more, so one foundation reaching both ends is our own test of the cross-task-family reuse claim. This line is in forward-looking R&D.
Characters and sets hold across shots — no restarting every take
Motion under physical constraints: no drift, no clipping
Script to final cut to multilingual release, one agent pipeline
R&DForward-looking stage
One foundationShared with the three shipping lines
GenerativeTask-family extension
NEXT FRONTIER
Language models have erupted.
Spatial intelligence is erupting now.
After language, AI's next door is understanding the 3D physical world. From 2025 to 2027 we work on one thing — machines that truly perceive space, reason about cause and effect, and interact with precision.
- 3D SPATIAL UNDERSTANDING
From 2D pixels to 3D modelling — reconstructing a scene's geometry, materials and structural relationships.
- PHYSICAL-WORLD REASONING
Infer how and why objects move, reasoning about cause and consequence by the laws of physics.
- EMBODIED PRECISE INTERACTION
Turn understanding into action — guiding robots to micron-level interaction in the real world.
Illustrative visualisation — a live point cloud once a camera is connected.
CUSTOMERS
Evidence from real production lines.
Multinational and tier-one manufacturers vet inspection suppliers hard; once you are in, it turns into multi-year repeat business and plant-to-plant replication. Every credential and case below is backed by a certificate or a signed contract.
Granted patents
32
including 15 invention patentsConcentrated in 3D measurement error compensation, structured-light calibration, depth-order reasoning, surface-defect detection and robot end-effector position compensation.
Software copyrights
96
National "Little Giant" specialised-and-sophisticated enterpriseAlso certified as a National High-Tech Enterprise and a Zhongguancun High-Tech Enterprise, with first-unit and Beijing new-technology product certificates.
MIIT exemplary case
285
one of 285 nationwide, 2025 AI application casesSelected for "industrial inspection agent built on a visual foundation model"; SkyVision was separately named one of 26 national industrial-AI scenarios.
Chemicals-major acceptance spec
0.1μm
grain size · 0.1% coverage · <1% repeatabilityDelivered for a leading international chemicals group: quantitative analysis of electron-microscope crystal images — three models trained from scratch in about two months, deployed inside the customer's own network.
FLAGSHIP CUSTOMERS · UNDER CONTRACT
SiemensBroseAdientGlobal chemicals majorMideaAUXInventecHebei ExpresswayPowerChinaAInnovationMegvii
ECOSYSTEM & PILOTS
FoxconnBYDCATLLuxshareChina MobileChina UnicomBAIC Mercedes-BenzHuaqinTsinghua UniversityChongqing University
PARTNERS
SIEMENSBROSENVIDIANGen CanadaInspurUnisplendour
THE MOAT
A data loop that compounds.
Peers do algorithms or hardware; we build the whole stack ourselves — from sensor and 3D imaging to the world model. One flywheel feeds all four product lines.
Real lines / camerasEvery inspection, every pick, every recognition
↓
Feedback flows backResults return to the DaoAI World foundation
↓
Stronger foundation → all four lines benefitAs the world model sharpens, AOI / robotics / SkyVision / Wemio sharpen with it
Why we're different
Feedback from real lines and real scenes is not something money can buy.
Eight years on plant floors, line by line, is engineering experience that accumulates slowly.
Full-stack in-house — from sensor to world model, we own the whole chain.
Hundreds of millions of real industrial samples, all sharpening the same world model.
On top of those real samples, DaoAI World amplifies training data 2–10× under physical constraints — being able to generate physically valid samples is itself proof that the model understands physics.
32 granted patents (15 of them inventions) plus 96 software copyrights, every one verified against its certificate.
Co-author of two industry standards: humanoid-robot perception and motion control, and general technical requirements for industrial robot vision systems.
FEATURED
Read deeper on physical AI
From the world model to the Series B — the path we took.
AI's next stop, the engine of physical AI: DaoAI World
WAIC 2026 turned toward physical AI: when AI enters the physical world, the real engine is a world model that reads three-dimensional space.
Read more
WeLinkirt closes its Series B: a homegrown world model for manufacturing
Xinhua reports: WeLinkirt closed a Series B led by the Chongqing Industry Guidance Fund, built on its own DaoAI World model.
Read more
Xinhua: WeLinkirt opens smart manufacturing's next stop with physical AI
Xinhua's "AI Insight": 2026 is the starting point for physical AI at scale — it has a verifiable commercial loop and real hardware to land on.
Read moreFAQ
What you might be wondering about us.
What does WeLinkirt DaoAI do?
WeLinkirt (DaoAI) is a physical-AI company. Its DaoAI World foundation model powers four product lines — industrial ACI inspection, robotic vision, SkyVision intelligent surveillance and Wemio content creation — all focused on making AI work reliably in the real physical world.
How is DaoAI's AI inspection different from traditional AOI?
Traditional AOI relies on hand-written rules or thousands of defect samples and per-board manual programming. DaoAI uses feature cognition from a visual foundation model: it models from a single good sample — no CAD, no defect-image library — and keeps learning from line feedback, getting sharper with use.
How long does it take to deploy a line? Do I need CAD drawings or defect samples?
The software auto-programs and goes live in about 5 minutes from a single good sample — no CAD drawings and no pre-collected defect library — and line operators can run it themselves.
Is the data secure? Does it have to run in the cloud?
No cloud required. DaoAI supports 100% on-premise (private) deployment, keeping line data on site, with flexible SDK / API / Docker integration to meet enterprise data-security and compliance needs.
What product lines does DaoAI offer?
Four product lines — industrial ACI inspection (ACI OS plus 2D and 3D ACI systems), robotic vision, SkyVision intelligent surveillance, and Wemio content creation — all built on the unified DaoAI World foundation model.
Which industries does DaoAI support?
Electronics / PCBA, semiconductor, automotive and parts, EV battery, pharma, food and agriculture, chemical, and consumer goods and general manufacturing.
GET IN TOUCH
Put physical AI into your scenario.
Manufacturing, city governance or content — we'd like to talk. Bring one good sample and we can build the model on the spot.