TabbyML

TabbyML is a company developing an open-source, self-hosted AI coding assistant called Tabby, which provides multi-line and full-function suggestions in real time. Tabby aims to enable developers to write code faster by offering suggestions based on context from comments and existing code. Tabby can be fine-tuned specifically to the user's project and is configurable, controlling the training process with YAML config. The team behind TabbyML aims to proliferate the use of AI-powered code assistants, targeting developers who may have legal, ethical, or governance concerns using products like GitHub Copilot. Users can host their own deployment or audit the entire open-source Tabby codebase for security or compliance.

Tabby is free and compatible with major coding LLMs, including CodeLlama, StarCoder, and CodeGen. The assistant offers optimizations across the stack, including IDE extensions (accurate streaming and cancellation with an adaptive caching strategy to increase the speed of completions) and model serving (parsing relevant code into tags to provide effective prompts). Tabby is self-contained without the need for database management systems (DBMS) or cloud services. It provides an OpenAPI interface that can be integrated with existing infrastructure and supports consumer-grade GPUs.

TabbyML launched in April 2023. The company was founded by two former Google employees, Meng Zhang and Lucy Gao. On April 6, 2023, an HN show launch introduced Tabby. The first stable release (v0.0.1) was introduced on August 31, 2023. On September 18, 2023, a new release enabled Tabby users to harness metal inference support on M1 and M2 Apple chips. Additionally, the team added support for the StarCoder series of models. On October 10, 2023, the company announced $3.2 million in seed funding from Yunqi Partners and ZooCap.