Kite launches Team Server to extend its code-completion ML assistant to enterprises

The ML-based code completion tool Kite launched its Team Server today, complementing its existing offerings for individual developers with an enterprise-grade version for entire teams of them.

Kite’s Team Server builds on its original coding assistant through deploying GPUs within enterprises’ internal networks, enabling them to create custom-trained models from their proprietary code to meet internal standards, and providing security features like its SSH tunnel proxy for encryption. According to Kite, this new server-scale ML model increases completions per line of code written by 40% more than its original, free desktop model. Team Server is also proportionally more complex, using 25 times more parameters — 100 million instead of 4 million — than its first iteration does.

Engineering teams have limited capacity, so Kite increases their capabilities. The tool frees their time from some repetitive tasks, including documentation searches and quick error fixes, by providing code snippet suggestions right in the code editor.

Now, Team Server’s advanced GPUs enable even higher-compute tasks. Its deployment model has two components: the containerized server itself, which is run on Linux and can be self-hosted on AWS, Azure, or GCP; and the desktop Kite client on each Team Server local machine. The Kite client includes a local engine that processes features like semantic completions and indexed local code. The client also has IDE plugins for delegating information to the source engine and an AWS companion app called Copilot that provides contextual documentation.

Enterprise IT teams that want to customize their Kite Team Server can put together and upload source code files to the server, train proprietary models on that source code’s particular textual patterns and for its internal standards, and activate these new models for its developers. Alternately, they can use the default general purpose model which suits most codebases.

Kite, founded in San Francisco in 2014, reports it now has over 400,000 developer users, including software engineers at 35% of Fortune 500 companies. Soon after Kite launched its private beta in 2016, it received 60,000 sign-ups and fielded dozens of inquiries from these companies as an on-premises solution. In 2017, Kite raised $ 17 million in a funding round that included GitHub CEO Nat Friedman, and it switched its ML engine offline from the cloud to reduce latency. Kite was initially only available for Python programmers, but last October it expanded to support 13 coding languages — now 16 in total, from Javascript to Typescript.

In a press release Kite provided, founder and CEO Adam Smith said, “We have grown from a single Python code completion tool into present-day where enterprise teams can use Kite. … Our users asked for this enterprise version of Kite. So this completions engine is a solution that is agile enough … to support thousands of simultaneous users.”