Glean Chat launches to unlock AI-driven search for the enterprise

Glean, the Palo Alto, CA-based enterprise search startup that hit unicorn status a year ago, announced Glean Chat today — a ChatGPT-like assistant specifically for workplace productivity. Users can answer questions and analyze information exclusively sourced from real-time information across the applications that make up their company’s knowledge base.

Glean’s efforts are based on a proprietary trusted knowledge model, which took four years to develop and is based on three pillars: company knowledge and context, permissions and data governance and full referenceability. In April, Glean introduced a suite of new features that use artificial intelligence (AI) to synthesize and surface relevant information from across an organization.

But unlike its original enterprise search offering, the Glean Chat user interface is conversational, like ChatGPT. The chatbot understands the thread and sequence of the conversation, so if you ask, “What’s the status of the Acme account?”, and then “Have they filed any support tickets?” and then “Who is their account representative?” and then “Which office are they in?”, Glean Chat will understand the sequential nature of the questions, so “their” means “Acme” and then “they” is the account representative.

Glean platform works across apps and knowledge sources

The company says Glean Chat is the only platform that works across apps and knowledge sources, including Microsoft 365, Google Workspace, Salesforce, Jira, GitHub and nearly 100 more applications. It also delivers answers that are personalized for relevance to each user, and adheres to all real-time enterprise data permissions and governance rules. Organizations can get up and running with Glean Chat in days.

“People are seeing the power of what ChatGPT can do, it’s no longer just about ‘help me find that document I’m looking for’ — people expect these AI assistants to do complicated tasks that they need to do on a regular basis,” Glean CEO Arvind Jain, who founded the company in 2020 after serving as a distinguished Google engineer, said in an interview with VentureBeat. “Now we’re not just the Google for your workplace but the Google and ChatGPT for your enterprise.”

That requires a two-step process, he explained, that first entails finding all the relevant information for whatever the user is asking in the search. Then, a large language model (LLM) is used to read all the documents and answer the question. That staged approach is what sets Glean apart from simply fine-tuning a large language model on a company’s data, he added.

Glean Chat preservers user permissions

Fine tuning an LLM works for certain use cases where data governance is not an issue, but “the problem you’ll run into is that you won’t be able to preserve all the permissions,” Jain said. “Let’s say I only have access to see this information — if some other person tries to issue the same question and query, then a model, which is only fine tuned once, will leak all of that information and that person might get an answer which they shouldn’t have access to.” To combat that, the model would need to be fine-tuned continuously, he explained.

Instead, he continued, Glean Chat is the “first offering that has been able to figure out how to manage this data governance layer in real time, and only show you things that you have access to, connect to all of the apps which you’re already using, and then [power] different workflows — sales workflows, support workflows, engineering workflows.”

Source