Kite, which suggests code snippets for developers in real time, today debuted integration with JupyterLab and support for teams using JupyterHub. Data scientists can now get code completions powered by Kite’s deep learning, which is trained on over 25 million open-source Python files, as they type in Jupyter notebooks.
Using AI to help developers is not an original idea. Nowadays you have startups like DeepCode offering AI-powered code reviews and tech giants like Microsoft working on applying AI to the entire application developer cycle. But Kite stands out with 300,000 monthly developers using its AI-powered developer environment.
Kite has been paving the way since its private debut in April 2016, before launching its developer sidekick powered by the cloud publicly in March 2017. The company raised $17 million in January 2019 and ditched the cloud to run its free AI-powered developer tool locally. In March, Kite launched a Pro plan, dipping its toes into monetization with a paid version. Kite comes from Adam Smith, who founded Xobni, an email service launched in September 2007 that Yahoo acquired in July 2013.
JupyterLab already offers native code completions, but they are sorted alphabetically and have no documentation. Kite offers longer ML-powered completions sorted by relevance, features over 100,000 Python docs for the highlighted completion, and does not require you to run a single cell in your notebook or press ‘tab’ to make completions appear.
Because Kite can complete multiple lines of code at a time, you spend less time writing boilerplate Python, such as import statements. It also learns and suggests your favorite aliases over time. Kite taps into Jupyter kernel completions so they show up automatically as you type. Kite’s models work locally and independently of your Python kernel, meaning if your kernel is busy reading in data, you’ll still get completions while coding in other cells.
Kite also includes additional features for JupyterHub teams:
- Deploy Kite on your JupyterHub server to bring Al-powered completions and one-click documentation to the whole team.
- Add Kite’s largest ML models to a GPU-powered server for smarter, longer completions.
- Custom-tailor Kite’s models to your team’s codebase and APIs.
- Manage Kite licenses and billing through a unified system.
To develop this plugin, Kite worked with the Jupyter community, JupyterLab’s development team, and Quansight Labs. The startup says it contributed to the Jupyter completions API and completions interface via four pull requests and 87 commits, making it easier to build plugins for JupyterLab in general.