Put an AI agent on every vacuum and semiconductor tool.
TARGON embeds a low-latency, task-oriented AI agent on the edge gateway beside your equipment. It reads the tool, diagnoses faults, predicts maintenance and tunes process parameters — and can run fully air-gapped, with no data leaving the fab.
Your best engineer's judgement is stuck in one person's head.
Vacuum and semiconductor tools are too complex for single-point PLC control. Fault diagnosis depends on the engineer who has seen that error code before; process know-how walks out the door when they retire or move lines.
Expert bottleneck
Complex working conditions mean pump selection and fault triage wait on a senior engineer — days, not minutes.
Fragmented response
Dozens of lines spread across sites, faults dispatched by phone roster, OEE drifting below industry average.
Data that can't leave
Process data is commercially sensitive. Cloud-only AI hits a hard compliance wall in fabs and research settings.
From a single tool to the whole enterprise.
A Local Industrial Agent in the workshop, and an Industrial Agent Plus at enterprise level — replicating the judgement of resident engineers and domain experts wherever it's needed.
Local Industrial Agent
A lightweight AI agent running on an edge gateway or industrial PC beside each vacuum or semiconductor tool. Low latency, task-oriented, and optionally offline for good.
- Air-gapped capable — data can be physically isolated
- Pre-loaded knowledge base — manuals, error codes, maintenance records
- Task-driven — diagnosis, guidance, predictive maintenance, process tuning
- Multi-agent coordination across tools and lines
Positioning: a device-and-process specialist that never resigns and works 24/7.
Industrial Agent Plus
Train a domain super-expert on your own private knowledge base. Serve customers on the outside, empower engineers on the inside.
- Private knowledge base — your know-how stays yours
- Dual mode — outward customer service, inward engineer support
- Continuous training — versioned rollout, explainable call chains
- Hybrid by design — sensitive data on-premise, general capability in cloud
Also available: Vacuum AI for pump lifecycle, and Smart Eye for quotations.
Deployed on real lines, with numbers attached.
Tier-1 vacuum pump OEM
Shanghai · Model computingComplex operating conditions meant pump selection depended on senior engineers — an average of 3 working days per proposal.
Vacuum AI built in the operating-condition surface and pump parameters to auto-match the optimal configuration.
8-inch wafer fab
Suzhou · Parts marketplaceTens of thousands of MRO spare-part SKUs, slow inventory turnover, purchasing premium, and emergency downtime relying on manual sourcing.
A parts marketplace aggregating suppliers, with AI predicting spare-part demand and auto-quoting.
Coating industry leader
Shenzhen · Service operations20+ lines across multiple sites, fault response dispatched by phone roster, OEE below industry average.
Local Industrial Agent plus a remote expert network — AI diagnoses first, then dispatches, with remote assistance when needed.
Questions engineers ask first.
What exactly is a Local Industrial Agent?
A task-oriented AI agent deployed on an edge gateway or industrial PC beside a vacuum or semiconductor tool. It embeds the tool's manuals, error-code library, maintenance history and process fundamentals, so it can diagnose faults, guide operators, predict maintenance and optimize process parameters on the spot.
Can it run without an internet connection?
Yes. It is built for air-gapped deployment. Data can be physically isolated from public networks, meeting the security and compliance requirements of semiconductor fabs, research institutes and industrial customers.
Which equipment and protocols are supported?
Industrial protocols including OPC UA, Modbus, SECS/GEM and MQTT. Built for vacuum pumps, coating lines and semiconductor fab equipment.
Why not just use a bigger model?
In industrial settings a smaller model plus a high-quality domain knowledge base usually beats a general-purpose large model. Process priors and engineer experience are the core asset — the model is just the carrier.
Talk about the machine that keeps failing.
A 45-minute technical call with an engineer. We'll assess your scenario and say plainly whether an agent is worth deploying.
Request a demo →