AI visual inspection that learns from your good samples.
Set up by your own operators. Deployed without touching your line. Running in production today on furniture-component, power-electronics, enamelware and window-manufacturing lines across Central Europe.
The Qi product family
Every product trains from good samples and runs itself — your line staff own the changeovers, not a vendor.
QiCHECK
AI visual inspection for assemblies and PCBs — presence, completeness, correctness and cosmetic defects. Multi-camera capable.
Explore the product →QiPACK
Packaging and seal inspection — verifies seals, part counts and labels so the wrong pack never reaches the customer.
Explore the product →QiSHELL
Closed-box inspection — labels, codes, print quality and a flawless package after sealing: no creases, no tears, glued right.
Explore the product →QiPICK
Neural-network picking — position and rotation of unorganized parts without a 3D scanner. Up to 90 picks per minute.
Explore the product →One team, four capabilities
AicobotiX is a Slovak AI & robotics engineering company. QiCHECK is the flagship — and it stands on everything else we do.
From good samples to a running line in four steps
Show good samples
Your operator trains detector zones per program in an afternoon. No labelling projects, no data-science team.
Run on the line
Every unit inspected, defects flagged instantly, every result logged with images.
Watch the dashboard
Pass rate, throughput and defects live — shift totals, hourly trends, full archive.
Scale
Roll out station by station, line by line, plant by plant.
Ask anything. QiCHECK answers.
QiCHECK-3 ships with an Analytics MCP server — the open protocol that connects AI assistants to live production data. Plug in Claude, ChatGPT or a fully on-premise LLM and ask in plain language. The answers come from your line, not from the model’s imagination.
Computed, not guessed
Numbers come from deterministic queries over the same production archive your dashboard uses — the AI only explains and charts them. That is why the answers always match the dashboard.
Every role, one question away
Plan burndown for the director, bottlenecks and line balance for production, defect Pareto with photos for quality, full serial-number trace for audits — and ad-hoc SQL for questions no report anticipated.
Claude, ChatGPT — or your own LLM
The MCP server is client-independent: connect the assistant your company already trusts, or an on-premise LLM so production data never leaves the plant.
Read-only, tokenized, audited
The service has no write operations. Per-user access tokens, every query logged, bound to your internal network — analytics, not control.
No internal LLM? We train your own
Without an internal LLM and the tools around it, most of your production data stays out of reach. We deploy an on-premise model and train your own LLM agent — connected to your data over MCP, tuned to your processes and terminology.
Real lines. Real dashboards. Live numbers.
Not lab demos — these systems inspect real production today.
Partners who stayed
Real professionals and partners in vision. One project meant inspecting transparent glue under freshly pressed glass with 0.1 mm tolerances — infinite variations, many constraints. We did it.
After years of searching for a system that would reliably check the output of our packaging line, AicobotiX delivered what even bigger companies with years of experience could not. A+++
From small production jigs to complex solutions — the sophisticated design, quality of workmanship and delivery times stand out.
Thanks to their competence and experience we could afford bold designs for our devices. A reliable and friendly partner.
Their enthusiasm, professional competence and serious approach helped us solve many difficult tasks together.
Engineering company from Trnava, Slovakia
AicobotiX s.r.o. builds on almost 20 years of industrial-automation experience. We develop our AI vision platform fully in-house and back it with engineering, prototyping and consultancy — for manufacturers from local SMBs to European blue-chip plants.
Start with a paid pilot on one of your parts
One line, one part family, 2–4 weeks. We report defect catch rate, false rejects and throughput — measured on your own production, not on promises.