Havells brings agentic AI voice control to 2.67 million mobile app users

Havells on Friday announced that it has launched an agentic AI voice feature on its Havells One app that lets users control connected appliances using natural speech in English and seven Indian languages.

The feature, built by Havells’ Center for Research & Innovation (CRI), is available to all 2.67 million users of the Havells One mobile app. It uses ElevenLabs’ streaming speech-to-text and text-to-speech models as the voice interface layer, integrated with Havells’ conversational intelligence and device-control agents.

The feature supports Hindi, Tamil, Telugu, Bengali, Kannada, Gujarati and Malayalam, alongside English.


Unlike Havells’ earlier voice-control experience, which supported English and Hindi on select appliances and required users to use specific commands, the new system is designed to interpret intent and handle natural and code-mixed speech.

For instance, a user could ask the system to “make the room comfortable”, prompting it to coordinate multiple connected devices such as an air conditioner, air purifier, fan and lights rather than requiring separate commands for each appliance.

The system can also execute requests across devices using different connectivity protocols, retain context across multiple interactions and ask users for clarification when a command is ambiguous. It can also explain device error codes in plain language.

The voice interface currently works across a range of connected Havells products, including fans, air purifiers, water purifiers and water heaters, as well as Lloyd air conditioners and washing machines, Crabtree smart switches and upcoming smart home products.

“Today, consumers just need a Havells Smart device registered in their Havells One App to use this feature,” said Dipesh Shah, executive president and CTO, Havells.

Shah described the feature as a daily-use AI experience that goes beyond traditional voice interfaces by allowing consumers to interact with devices without having to remember specific commands.

Havells said it evaluated voice AI models on recognition accuracy, multilingual support, latency and voice synthesis before moving the solution into production in approximately six weeks. ElevenLabs’ Forward Deployed Engineering team supported voice quality and regional language pronunciation tuning.

“People have always had to learn the interface of communicating: the commands, the syntax, the order they come in,” said Karthik Rajaram, general manager and country head, India, ElevenLabs.

The company said the challenge is particularly relevant in India, where households often switch between multiple languages in everyday conversations.

Havells plans to expand the feature to more Indian languages and bring additional appliances onto the platform. It also plans to add deeper personalisation for individual household members and extend the voice interface beyond device control to product FAQs, troubleshooting and scheduling.