Most AI assistants are built in English and then translated. Kruti wasn’t. It launched in 13 Indian languages from day one—and that distinction matters more than it sounds.

Krutrim is India’s first AI unicorn, valued at $1 billion and spun out of Ola by Bhavish Aggarwal. They launched Kruti on June 12th. It’s an “agentic” AI assistant, which these days gets slapped on anything that sounds fancy. But Kruti actually earns the label: it doesn’t just answer questions. It books cabs, pays bills, orders food, generates images. Real transactions through real APIs—not chatbot parlor tricks.

What caught my attention—from a DevRel perspective at Google—is the voice-first design. India has 1.4 billion people and extraordinary linguistic diversity. A text-first, English-default interface? That excludes most of the population. Kruti’s voice-first approach supports Hindi, Tamil, Telugu, Bengali, Marathi, Kannada, Gujarati, Malayalam, Punjabi, Odia, Assamese, Urdu, and English. It reflects a product decision global AI companies keep getting wrong: you build for the market, not for the demo.

The dual memory system is technically interesting—short-term for session context, long-term for personalization across sessions—though not exactly novel. What’s actually new is applying it where users switch languages mid-conversation. This happens all the time in multilingual households. Western AI assistants handle it poorly. Kruti doesn’t.

Kruti runs on Krutrim’s V2 large language model alongside open-source models. Agents pick the appropriate model based on task context. The developer SDK is already out for third-party integration, with memory management and tool orchestration hooks. Smart move. Platform value compounds with ecosystem participation; opening the SDK early signals Krutrim wants developers building on Kruti, not just using it.

I’m watching this closely. It represents something the global AI market needs more of: products built from scratch for non-English, non-Western contexts. Not adapted after the fact—built for them. The voice-first, multilingual-native, task-completing architecture isn’t just localization strategy. It’s a different product philosophy. And it’s one a billion people have been waiting for.