Introducing Refact for Enterprise
Back in May, we released an open-source self-hosted version of Refact.ai which has reached almost 1000 stars on GitHub.
While our self-hosted solution has been popular among individual developers, we recognized that there’s a set of requirements for the companies that are not relevant to individual devs, but are a must-have in an enterprise environment.
That’s why today we’re releasing Beta of Refact.ai Enterprise Edition which still has the same great developer experience as in the self-hosted version, but has additional functionality for access control, security, and even faster inference.
With Enterprise edition of Refact.ai you can easily get a local AI coding assistant fine-tuned to your codebase and working blazingly fast for all the developers in your company, while scaling your infrastructure.
Refact.ai Enterprise server, running chat and completion models on 8 GPUsFaster and more powerful inference
In addition to Refact.ai own 1.6B code model, Enterprise Refact.ai supports the most powerful open-source code models, which include StarCoder, WizardCoder, Code Llama, and more.
On top of that, we have added PagedAttention and continuous batching support to our inference engine by using the VLLM library. With these improvements, you are likely to see a 4-5x increase in total throughput for the same hardware!
Companies who prefer a hybrid option with both self-hosted and cloud models can do so by connecting OpenAI API.
Automatically fine-tune models on your codebase
With Enterprise Refact.ai, you have a company-wide hub for creating your own fine-tunes of code models that don’t require any special knowledge of LLM fine-tuning. Your source code never leaves your local environment and you have full control over which fine-tune model to use. The whole process of fine-tuning takes a couple of clicks.
Compared to a base model that didn’t train on your codebase, a fine-tuned one remembers APIs, style, tricks, and formatting typical for your company, thus increasing the quality of suggestions.
Admin Control and enterprise-grade security
Enterprise Refact.ai adds additional admin controls to specify which users and teams have access to the models, and how GPUs are used.
Refact Enterprise is safe to expose to the internet, protected by passwords and unique API tokens for each user (please read the manual) and talk to your admin about how to do it correctly). This allows the use of Refact.ai Enterprise remotely even without an always-on VPN.
Get Code Suggestions inside your favourite IDE
With the support of the most popular IDEs such as VS Code and JetBrains, developers can easily connect local models and get code suggestions in the IDE of choice.
We also have plugins for Visual Studio, Sublime Text, and VIM to offer developers even more choices.
If you’re interested in trying Enterprise Refact.ai for your company, contact the team and we’ll provide you with the license.
