GitLab has expanded its AI lineup with Duo Chat

Earlier this year, GitLab exposed Duo, a suite of AI features that aim to help developers be more productive by summarizing issues and creating descriptions of epics and issues, as well as through code suggestions and vulnerability explanations, and other features. Now, the company has added Duo Chat to this lineup, a ChatGPT-like experience that allows developers to interact with the bot to access existing Duo features, but in a more interactive experience. Duo Chat is now in beta.

During an interview at KubeCon earlier this week, David deSanto, GitLab’s Chief Product Officer, told me that the idea here is to transition many of Duo’s existing capabilities to chat between this first beta and launch. in GA.

Image Credits: GitLab

DeSanto, who has increasingly become the company’s public face, also noted that GitLab has moved its chat backend to Anthropic’s Claude. That may come as a bit of a surprise, since GitLab has partnered with Google for other AI features, including its code completion service. “We didn’t pick (a model) for the chat,” deSanto told me. “We used our own model – an open source model – and we found that based on the way chat works within GitLab and the large context that is required, a 100k context window makes it much easier to get better information in and out.”

He also noted that GitLab is partnering with Oracle for cloud-based GPUs. “We are cloud-agnostic. We look for the best and all three (Google, Oracle and Anthropic) really value each other through us, so it’s a good relationship.

Currently, in addition to more general code-related chat capabilities, features such as ‘explain this code’ and code refactoring are available in Duo Chat. Test case creation, vulnerability explanation and other features will follow soon.

DeSanto noted that over the course of testing these features, it became clearer how important context is for getting the best results from this chat experience. An advantage of GitLab over other players in this field is that it can access all your code, even if it is not open in the IDE. Additionally, these large context windows make it easy to track past conversations.

An interesting observation of the team while watching developers use the tool during the closed preview period is that experienced developers don’t always like the code suggestions because they can be noisy. Instead, they prefer to access these AI tools through Duo Chat.

“I thought chat would be more for fellow developers but they seem to prefer inline code completion — and then older developers are more and more drawn to doing code in chat,” deSanto said. . “Noise is the cause. They said: I know what I’m typing. I don’t need you to tell me. But maybe they want some guidance or some refactoring, for example.

Leave a comment