Zeli srl, an Italian distributor of solenoid valves, hydraulics, and industrial automation components based in Modena, has deployed an LLM-powered chatbot called Liara as the primary customer-facing technical support interface on its web platform. Rather than treating the AI assistant as a supplementary feature, Zeli has made it the front door for technical queries, fielding questions about product selection, valve specifications, and vendor availability across its catalog. The company shared the deployment on Hacker News under the framing of a broader architectural question: when should an LLM be core infrastructure rather than a bolt-on tool?
Liara operates with several deliberate constraints that reflect the realities of running inference in production without authentication. Conversations are stateless with no persistence between sessions, messages are capped at 500 characters, and a rate limit of 3 messages per minute is enforced. These measures manage inference costs and limit abuse. The system runs on EU datacenter infrastructure with native GDPR compliance, and a separate "On-Premise" label suggests either a hybrid hosting model or an on-premises inference option aimed at industrial customers with data sovereignty requirements. Zeli also includes an explicit accuracy disclaimer directing users to call their support line for critical decisions — a pragmatic hedge given the stakes of component selection in industrial automation.
The vertical specificity of the domain may be a key factor in Zeli's confidence to deploy Liara as a primary interface. Solenoid valves and industrial automation carry well-defined technical vocabularies, constrained product catalogs, and relatively predictable query patterns. Those conditions make domain-specific training far more manageable than in open-ended consumer applications. The company describes Liara as under continuous training, suggesting the system is being refined against real user interactions. Whether that can maintain the reliability industrial customers expect is the open question the Hacker News thread is probing.
For SME distributors, the economics of technical support push hard toward automation: product knowledge is specialized, staff time is expensive, and many customer queries are repetitive. An always-available AI assistant has obvious operational appeal. Zeli's implementation, cautious disclaimers and rate limits sitting alongside its central placement in the customer experience, shows the tension between ambition and reliability that defines early agent deployment in industrial settings. Whether Liara holds up under that weight is something the Modena company will find out in production.