Programming with AI continues to be a extremely divisive subject, however there’s no denying that increasingly builders are beginning to incorporate AI into their day by day workflows. Whether or not you’ve already picked your facet within the debate or are nonetheless undecided, we’ve received a brand new function in v2024.1 of JetBrains IDEs that may simply pique your curiosity – full line code completion. It’s AI-powered and runs regionally with out sending any knowledge over the web.

On this weblog submit, we’ll let you know extra about what full line code completion is, the way it works, what languages are supported, and how one can present suggestions about it to us.

What’s full line code completion in JetBrains IDEs?

This new kind of code completion was added to JetBrains IDEs with the most recent 2024.1 replace. As you may see under, it takes the type of gray-toned, single-line options that full traces based mostly on the context of the present file:

These options are powered by specialised language fashions that we’ve skilled particularly for various languages and frameworks. The fashions run regionally with out sending any code over the web.

Full line code complеtion is at the moment out there for Java, Kotlin, Python, JavaScript, TypeScript, CSS, PHP, Go, and Ruby throughout the corresponding JetBrains IDEs: IntelliJ IDEA Final, PyCharm Skilled, WebStorm, PhpStorm, GoLand, and RubyMine. Within the coming months, we plan to increase the performance to C#, Rust, and C++, so it can additionally land in Rider, RustRover, and CLion.

Notice that full line code completion is included along with your energetic JetBrains IDE subscription at no extra price – simply ensure you’re on v2024.1 or later. If you happen to don’t but have a subscription, you can too use this function throughout the 30-day free trial.

How does full line completion work?

With full line code completion, we had two major objectives in thoughts. The primary one is clear – that will help you save time and improve your coding pace. However past that, we additionally wished to supply an answer that addresses the basic constraints of cloud-based options and places the privateness of your knowledge first.

Right here’s a breakdown of how full line code completion helps to comprehend these two goals:

• It really works regionally and is on the market offline. This implies you may reap the benefits of the function even if you happen to aren’t related to the web.
• It doesn’t ship any knowledge out of your machine over the web. The language fashions that energy full line code completion run regionally, which is nice for 2 causes. First, your code stays protected, because it by no means leaves your machine. Second, there are not any extra cloud-related bills – that’s why this function comes at no extra price.
• It’s built-in deeply into JetBrains IDEs. All options might be appropriately formatted, with the IDE checking for balanced brackets and quotes. Moreover, we use the ability of static evaluation and our understanding of code to filter out incorrect options. Every supported language has its personal set of recommended code correctness checks. Essentially the most fundamental ones, like unresolved reference checks, are applied for many languages to ensure that the IDE doesn’t counsel non-existent variables and strategies. The auto-import function can also be supported.
• It’s designed to maintain your workflow as easy as potential. We use good filtering to keep away from exhibiting options that are typically canceled explicitly or deleted proper after they have been added.

For some extra technical particulars, see this part under.

Full line code completion vs. AI Assistant

There are two methods you may profit from AI performance in JetBrains IDEs – full line code completion and JetBrains AI Assistant. We admire that this could be complicated, so let’s take a better have a look at what they’ve in widespread and the way they differ.

Each full line code completion and JetBrains AI Assistant intention that will help you work quicker. They each additionally transcend the usual completion that has been out there in JetBrains IDEs for a while already. Nonetheless, JetBrains AI Assistant is a separate product in its personal proper. Its function set isn’t restricted to code completion – it provides you lots of different issues, too, together with context-aware good chat and the flexibility to generate exams or write documentation.

See the desk under for a comparability of the 2 AI functionalities:

Please relaxation assured that we by no means practice any of our AI options on prospects’ code. If your organization has strict knowledge privateness rules, however you continue to wish to pace up your workflows with AI, full line code completion could also be a more sensible choice for you.

Below the hood

The spine of full line code completion is a programming-language particular language mannequin, which is skilled in home utilizing a dataset of open-source code with permissive licenses. The language mannequin’s enter is the code earlier than the caret, although for some languages, we additionally add content material from associated information. The output is the mannequin’s recommended continuation of the present line, which is proven in grey.

The language mannequin’s inference runs in your native machine. To make sure essentially the most environment friendly era, the mannequin inference runs in a separate course of and is closely optimized for the goal machine’s structure. For instance, if you happen to’re utilizing x86-64 structure, the mannequin will run on the CPU, whereas if you happen to’re utilizing ARM64 structure, the mannequin will use the ability of your laptop’s GPU.

After the suggestion is generated, plenty of post-processing steps are utilized. First, we verify whether or not this suggestion is syntactically and semantically appropriate, after which we carry out good filtering, formatting, parenthesis balancing, and varied different manipulations. Publish-processing is essential for consumer expertise, so we do our greatest to indicate solely worthwhile options that don’t disturb your workflow.

Lastly, you might also be questioning why we determined to go for single-line options. The size of the AI completion options is a trade-off. Whereas longer options do have a tendency to scale back what number of keystrokes you must make, which is nice, in addition they improve the variety of evaluations required in your finish. Taking the above under consideration, we determined that finishing a single line of code can be a good compromise.

This determination allowed us to scale back the scale of the mannequin with none important decline in suggestion high quality. Within the 2024.1 model of JetBrains IDEs, we use a language mannequin that has 100 million parameters, with a most context dimension of 1,536 tokens, which is roughly 170 traces of code.

How you can tweak the function

You’ll be able to configure full line code completion in Settings | Editor | Normal | Code Completion – all of the settings may be discovered there, underneath the Machine Studying-Assisted Completion part:

If you happen to’d like to show off the function, you are able to do so by unticking the Allow Full Line options checkbox. Alternatively, you may disable the plugin powering this function. To take action, go to Settings | Plugins, swap to the Put in tab, and search for full line code completion.

How you can present suggestions

Full line code completion continues to be in energetic growth, so we encourage you to share your suggestions with us. You are able to do so by leaving a remark underneath this weblog submit. You may as well upvote present points right here or create a brand new one by logging in and clicking on the New Challenge button within the prime right-hand nook.

That’s it for in the present day. Please give full line code completion a strive and tell us what you suppose. We’ll proceed bettering this performance additional, with help for C#, Rust, and C++ in addition to higher integration with AI Assistant’s multi-line code completion being our prime priorities for now. Keep tuned for updates!

Subscribe to JetBrains Weblog updates