Tabby is a self-hosted AI coding assistant that runs on your own infrastructure as an on-premises alternative to GitHub Copilot. The server is self-contained, with no database management system or cloud service required, exposes an OpenAPI interface for integration with existing tooling such as cloud IDEs, and runs on consumer-grade GPUs, including Apple Metal on M-series hardware.
You start a server with a chosen completion model and chat model, for example StarCoder-1B for completion and Qwen2-1.5B-Instruct for chat, then connect editor extensions for VS Code, Vim, and JetBrains IDEs. Completion is retrieval-augmented: the server indexes your repositories and pulls in repo-level context, declarations from the local language server, and recently modified code to shape suggestions.
Alongside inline completion, the Answer Engine gives engineering teams a knowledge base over their own code and documentation. It answers questions in the chat side panel or the web UI, can switch between backend chat models, turns conversations into persistent shareable pages, and indexes GitLab merge requests and custom documentation added through REST APIs. Team features include an admin UI, reports with team-wise usage analytics, GitHub and GitLab integration, LDAP authentication, and single sign-on.
Features
- Code completion: inline suggestions with multiple choices, informed by repository context, LSP declarations, and recent edits
- Chat and inline edit: a side-panel chat with an @ menu for referencing files, plus edit-via-chat and right-click inline editing in VS Code
- Answer Engine: question answering over internal code and docs, with shareable pages and switchable chat models
- Data connectors: self-hosted GitHub and GitLab repositories, GitLab merge requests, and documentation ingested through REST APIs
- Model registry: CodeLlama, StarCoder, CodeGemma, CodeQwen, Codestral, and Qwen chat models, selectable at launch
- Deployment options: Docker, llamafile, and SkyServe on any cloud, with CUDA or Metal inference
- Team administration: admin UI, usage reports, activities page, LDAP, GitLab SSO, and secured access
- Editor extensions: VS Code, Vim, and JetBrains, with auto-generated commit messages
- Pochi agent: a companion VS Code agent that takes GitHub issues to pull requests with CI, lint, and test results in the sidebar