Refact aims to make code generation AI more attractive to enterprises

Image credits: Tippapat/Getty Images

In 2021, Oleg Klimov, Vlad Gubar and Oleg Kiyasko began jointly developing the platform Refact.ai. This will give users more customization and control over their experience, and will help convince more companies to adopt his GenAI for coding.

Klimov and Kiyasko have been collaborating on building AI-based systems for image recognition and security systems for nearly a decade. Guber had known Kiyashko since childhood. They were neighbors in the southern Ukrainian town of Yuzhnukrainsk.

“It was clear that AI was going to change the very idea of ​​what engineering is,” Klimov told TechCrunch in an email interview. “As software engineers at heart, we decided we needed to be in the best position to get through this by building an independent system for software engineering.”

Most developers are aware that AI is revolutionizing their profession. 82% of respondents to his recent HackerRank poll said they believe AI will “redefine” the future of coding and software development.

Refact leverages generative AI models trained on allowed code to power code suggestion platforms for developers.

The majority has embraced this change, with VC firm HeavyBit’s 2023 survey finding that 63% of developers say they use GenAI for coding tasks. But employers are even more skeptical. In a separate survey of business executives and IT professionals, 85% expressed concerns about GenAI’s privacy and security risks.

Companies such as Apple, Samsung, Goldman Sachs, Walmart, and Verizon have gone as far as restricting internal use of GenAI tools due to fears of data breaches.

So how is Refact different? Klimov says it runs on-premises.

Similar to GitHub Copilot, Amazon CodeWhisperer, and other leading GenAI coding assistants, Refact answers natural language questions about your code, such as “When was this dependency last updated?” It can recommend lines of code and tweak them to improve performance on your specific codebase. .

“One way to think of it is to think of him as a ‘talented young engineer’ or an artificial colleague on a team that is highly productive but needs supervision,” Klimov said.

However, unlike many if not most of its competitors, Refact does not require an internet connection. It doesn’t even upload basic telemetry data, Klimov claims.

Refact can be run offline, on-premises, or in a cloud-hosted management setup.

“We recognize the challenges around data sources and use, security and privacy, and are developing better controls and processes. [enterprises] We want to face and ensure the integrity of our customers’ information and innovative breakthroughs,” Klimov said.

Powering Refact’s platform is a compact code generation model trained on permissively licensed code, which Klimov claims is also a key competitive advantage. Some code generation tools trained using copyrighted or restrictively licensed code have been shown to regurgitate that code when prompted in certain ways. and (at least according to some IP experts) pose potential liability risks for implementing companies.

Vendors like GitHub and Amazon have introduced settings and policies aimed at allaying the fears of companies wary of intellectual property challenges with GenAI coding tools. However, it is unclear whether they have made much progress. In his 2023 survey of Fortune 500 companies conducted by Acrolinx, nearly one-third said intellectual property was their biggest concern regarding the use of generative AI.

“We used permissive license codes for training [our models] Because our customers asked for it,” Klimov said.

Refact’s privacy and intellectual property-conscious approach has led to it raising $2 million in funding from private investors and delivering approximately 20 net pilot projects with enterprise customers. The platform, which is also available in cloud-hosted plans starting at $10 per seat per month, is generating revenue and is currently on track to generate “millions” annually by this summer, according to Klimov. .

This is surprising considering that vendors like GitHub have a hard time making money with their code generation tools. Copilot was reportedly costing GitHub’s parent company, Microsoft, up to $80 per user per month as a result of the associated cloud processing overhead.

Refact is one of a growing cohort of GenAI tools for programming.

The near-term focus of the London-based Refact team of eight people is to upgrade Refact to be able to run code autonomously, perform “multi-step” plans, and self-test code.

“We are actively working on the next generation of AI assistants that can debug the code they write and work with any large codebase,” Klimov said. “We are well-funded internally and have the capital we need to continue developing our products. But what really benefits us is the availability, and the enthusiasm of the very talented people who are joining the AI ​​revolution, and the ability to make Refact a growing and lasting company. We see it as a place to develop something that can have an impact.”