Professional software development tools company JetBrains s.r.o today announced the release of a new artificial intelligence-powered coding companion called “AI Assistant,” now generally available to users of the company’s independent development editors.

The new coding assistant is powered by multiple generative AI models and deeply integrated with the company’s IDEs, which are software applications used by software developers and engineers that combine the tasks of writing code, building, debugging and testing into a single platform.

By integrating the AI Assistant into the IDE, its models can understand both the code and the overall project without the need for the developer to switch out of their editing environment to interact with it. The assistant provides a conversational interface with a large language model where the developer can ask questions, request code review, ask for the generation of code snippets, and then easily copy/paste code directly into the project.

Jodie Burchell, developer advocate for data science at JetBrains, told SiliconANGLE in an interview that what differentiates the AI Assistant is that under the hood, it’s powered by an AI Service that transparently connects it to the right LLM provider based on developer needs.

“This is basically a service that is contacted by a network within the IDE. Right now, we determine heuristically which model to use based on which one is going to be the best for the job,” Burchell said. “In the future, we may introduce the capability to choose the models that you want to work with. And this is potentially not that far off as we are considering allowing organizations to bring their own models.”

Currently, JetBrains uses a combination of OpenAI’s latest models through the AI Service including GPT-3.5 and GPT-4 in addition to its own in-house LLMs. Burchell added that the company also recently signed on with Google and will soon be integrating some of the company’s models as well.

“Generally, the idea will be, if it’s really important in terms of privacy or fine-tuning, you may have the opportunity to bring your own model,” Burchell added. However, this will naturally come with a bit of a compromise as some models will not be as powerful as others. “So, this is something we’re working on, but it will not be part of the initial release.”

Most important, Burchell stressed, JetBrains has been very careful about agreements with AI providers that no data will be stored or used for training. This means that any company that uses the AI Assistant can be certain that their data will remain private.

From within the editor, the capabilities of the AI Assistant include multiline code completion, assisting with documentation, and test generation. Developers can also use the chat interface to ask the assistant to explain their code and have it suggest identifiers and function names based on the entire context of the project. Developers can save time by using the AI to refactor code or change their codebase programming language from one to another.

As with any AI tool, the AI Assistant can still produce errors or hallucinate, which Burchell said is greatly reduced by providing the LLM as much context as possible and making sure that its information is as up-to-date as possible. To help reduce hallucinations, JetBrains is looking into a process called retrieval-augmented generation, or RAG, which uses real-time data to maintain the knowledge base of an LLM to reduce the number of errors it generates.

“This is also the reason we chose AI Assistant,” said Burchell. “We really do want to emphasize that we’re dealing with models that are fluid and they can hallucinate. You should not be just accepting things at face value; you should use other functionality in your IDE to help refine code and own it. We don’t want people to think this is going to replace them. It’s just another productivity tool.”

The new AI Assistant plugs directly into JetBrains’ commercial IDEs and will eventually be available across the company’s entire fleet of products. It’s priced at $10 per user per month for individuals, or $100 yearly. For organizations, it’s $20 per user per month or $200 yearly.

Image: geralt/Pixabay