TabbyML Inc., a startup with an artificial intelligence productivity tool for developers, has raised $3.2 million in seed funding from Yunqi Partners and ZooCap.

TechCrunch reported the investment today. Meng Zhang and Lucy Gao, the startup’s two co-founders, told the publication that the newly raised capital will be used for product development.

TabbyML offers an AI tool called Tabby that enables developers to generate code using natural language prompts. The tool is designed to be integrated with the code editor that a software team uses in its day-to-day work. Within the editor’s interface, a developer can type in a sentence describing a task and have Tabby automatically write software capable of performing that task.

The company says the AI models powering its tool can generate multiple lines of code at once. Moreover, they’re capable of writing entire functions, which are units of code that each perform one specific task. That one task can vary significantly in complexity: some functions are written to perform simple arithmetic operations, while others carry out complex computations on large volumes of data.

TabbyML positions Tabby as an alternative to Microsoft Corp.’s GitHub Copilot service. Introduced in 2021, GitHub Copilot is a cloud-based coding assistant designed to help developers write software faster. Like Tabby, it allows users to generate code with natural language prompts.

One of Tabby’s main differentiators is that it’s open-source, which means it can be used for free. GitHub Copilot is priced at either $10 or $19 per user per month depending on the edition. Additionally, Tabby’s open-source license allows companies to customize its code for their requirements.

TabbyML says that its tool can also be customized without any code modifications. A fine-tuning feature included out of the box enables developers to train Tabby on files from the software project they’re working on. The contextual information the tool gleans from the project enables it to generate more relevant code suggestions.

Besides Tabby and GitHub Copilot, there are several other AI tools on the market that are capable of automatically generating code. The large language models that power those tools contain more than ten billion parameters in some cases. Because of their complexity, such models can only run on expensive data center graphics cards.

TabbyML says it has taken a different approach. According to the company, the models powering its coding assistant have between one and three billion parameters. That means they’re compact enough to run on a desktop computer equipped with a consumer-grade graphics card.

The startup distributes Tabby in the form of a downloadable Docker container. Developers don’t have to install external components such as a database to use the tool. Additionally, the company says that there’s no need to connect to a cloud-based backend, which means Tabby can work without an internet connection. 

Tabby has received more than 11,000 stars on GitHub since its release in late August. Stars are used by GitHub users to bookmark open-source projects of interest and are a measure of developer adoption. 

Image: Unsplash