Glean Pilots Real-Time Voice Assistant Powered by OpenAI
Enterprise search company Glean has announced a beta for its Assistant's real-time voice feature. The tool enables sub-second latency conversations over enterprise data sources like calendars, CRM, and codebases. It is powered by OpenAI's GPT-realtime-1.5 API and is designed to operate without sending enterprise data externally.
- Glean's CEO and co-founder, Arvind Jain, was previously a Distinguished Engineer at Google for over a decade, where he worked on search, and was also a co-founder of the cloud data management company Rubrik. - The company's core technology is not based purely on vector search; it employs a hybrid retrieval strategy combining traditional information retrieval signals (like BM25) with modern vector search. This architecture is built on top of an "Enterprise Graph" that maps relationships between people, content, and activities to enforce access controls before data is passed to a large language model. - To address enterprise security concerns, Glean utilizes a single-tenant architecture, meaning each customer gets their own dedicated cloud project. The platform is designed to be secure from the ground up, following a zero-trust model and holding SOC 2 Type II and ISO 27001 certifications. - The underlying OpenAI model, gpt-realtime-1.5, is a native speech-to-speech engine, which avoids the latency of separate transcription and text-to-speech steps. OpenAI reports that this new model improves alphanumeric transcription accuracy by over 10% and instruction following by 7%. - Glean has raised a total of $765 million and recently secured a $150 million Series F round in February 2026, leading to a $7.2 billion valuation. The company surpassed $100 million in annual recurring revenue as of early 2026. - Achieving sub-second latency in voice AI is a significant engineering challenge, with total response time comprising speech-to-text, LLM time-to-first-token, and text-to-speech generation. Production-grade agents aim for latency below 800 milliseconds to feel natural to a user. - The voice feature is an extension of the Glean Assistant, which already performs tasks across enterprise applications like Jira, Salesforce, and GitHub through more than 100 available actions. The assistant creates a personalized "personal graph" for each user by learning from their activities, chats, and documents to tailor its interactions.
