a selection of various AI tools for developers

a selection of various AI tools for developers

Behind the well-known GitHub Copilot lies a huge world of useful applications for programmers, and every week something new appears in it. In this post, we will talk about these tools, both full-fledged competitors of the GitHub product, as well as more specific plugins, as well as our own development in this direction.

Cody AI

Expansion Cody AI created by the same team that made SourceGraph. Cody AI loads your repository into the SourceGraph database and relies on it every time it answers a question. The tool supports many functions called “recipes”. For example, you can highlight text and ask for a JS-style comment to be written to it. Or ask to “sniff” the selected piece of code – and in response you will receive a list, which will include not only the necessary fixes, but also points “to think about”. For example, a large number of if in the code; if you don’t like it, you can ask to optimize the code. In response, you will not only be offered an alternative, for example, with switch, but also told why it will work faster.

SourceGraph is a paid product, but Cody AI is free; Now, in the early stage of the product, the only limit is the number of requests to GPT 3.5, 250 per month. Cody AI is economical in requests and skilfully compresses the context, so the limit is not so small.

Amazon CodeWhisperer

CodeWhisperer – an analogue of Copilot, created by Amazon and almost as good as it. But at the same time it is free. CodeWhisperer is constantly evolving, and if you’re new to AI tools at work, this is a good place to start before you commit to another line of spending.

Copilot Labs

Copilot Labs – A free extension for VS Code that is separate from Copilot itself. It is based on roughly the same recipes as Cody AI, but here they are called brushes. A remarkable possibility of setting up custom “brushes” — scenarios to which you can regularly return in your work. They can include various steps — fixing bugs, creating lists, improving readability, etc. building In addition to custom brushes, Copilot Labs can create tests, explain code, and translate it into another language.

work example

Lintrule

Lintrule – This is an LLM (large language model) that conducts code reviews. It works through the command line. Lintrule can upload config files to work and take them into account during review. This is not a cheap tool, a review of 1000 lines of code will cost $1.

Lintrule knows how to find things in the code that ordinary linters will not notice. For example, Lintrule knows what personal information is and will track whether it is stored somewhere in your logs, potentially violating the privacy policy.

OpenCommit

OpenCommit is a commit tool that won this year’s GitHub hackathon. It sees all your code changes and outputs a short, formatted description of what you’ve done. That way, before committing, you don’t need to commit all the changes from the beginning — it’s a good time saver. OpenCommit can be integrated with GitHub Actions.

An example of a commit before and after OpenCommit processing. Source

OpenCommit works through an OpenAI API key and uploads only changes there, so the cost of using the tool is small – on average a few dollars per month.

Replit Ghostwriter

Ghostwriter is a product of Replit, the creator of the browser IDE and its own LLM with open source code. Ghostwriter is only compatible with their own IDE, where it takes over the functions of Copilot. The project will be of interest primarily to those engaged in machine learning, as the Replit IDE integrates projects from the AI ​​community HuggingFace – tens of thousands of different models, datasets and ready-made applications. Among them, StarCoder stands out — a model for writing code with 15.5 billion parameters, trained in more than 80 programming languages.

CodeGeex

CodeGeex — an alternative to Copilot from Chinese developers, which can become an important factor in the current environment. CodeGeeX can generate code, translate it, add comments, fix bugs, explain code and more. Available for free as an extension for popular IDEs. For issues of corporate use with priority support and deployment in the company circuit, you can write to the developers.

Codium AI

Codium AI focuses exclusively on creating tests. Distributed as a plugin for VS Code and the JetBrains IDE, it supports only Python, JS and TS so far, but they promise to add Java soon. Also in the arsenal of developers is an AI tool for pool requests, and in the future it is planned to introduce a full-fledged platform for coding.

Tabnine

Tabnine – Another extension on our list that offers good functionality. Tabnine can help with refactoring and linting, automatically generate documentation, and add code. Distributed for free as a plugin for popular IDEs.

CodeT5

CodeT5 intended primarily for code generation from descriptions given in natural language. In addition, the tool can translate and make short descriptions of the code to make it easier to understand.

OpenAI Codex

The selection would be incomplete without a mention Codex from OpenAI The main function of this tool is the translation of natural language into a programming language, but Codex offers many other functions on the basis of which enthusiasts are already making their tools.

PolyCoder

PolyCoder – Codex competitor, fully open to community development. You can add and improve its datasets yourself. PolyCoder was created with ChatGPT 2 in mind, but the creators say it’s still the best AI tool for C development.

Snyk Code

Snyk Code – Another relatively narrowly focused product. Snyk Code does not add code, does not automate tasks, but conducts static tests of application security. And because it happens in real time, you don’t need to analyze the reports separately. All popular IDEs and programming languages ​​are supported, there is a free plan for several hundred tests per month.

GigaCode

Finally, we will mention our product. We recently introduced GigaCode, an LLM platform that we are developing in parallel with GigaChat. GigaCode is able to generate a continuation of the code by its fragment, conduct a semantic search, look for some vulnerabilities. Several thousand of our developers are currently using GigaCode; we are planning a public release in the coming months, but for now it is in limited beta. GigaCode supports Java, JS, TS, Python, C/C++ and a number of other languages, as well as popular development environments – IDEA, PyCharm, VSCode, Jupiter and others.

Related posts