Biscuit: Native AI Code Editor, Under 20MB

Biscuit is a fast, native code editor designed for speed and extensibility, weighing in at less than 20 MB and installable in seconds via pip. Installation is straightforward: running pip install biscuit-editor followed by biscuit path/to/src opens a project immediately. The editor bundles several AI features powered by LLM agents; it supports Gemini (2.5 Flash and Pro variants) and Anthropic Claude (4.5 Opus, Sonnet, and Haiku). The planning agent can perform a range of file and codebase operations: reading and editing files, deleting files, listing directories, searching with glob patterns and grep, searching the codebase directly, running terminal commands, writing to-do lists, and inspecting workspace and active editor state. Users can also attach files to add context in chat interactions. The editor handles code intelligence through fast tree-sitter–based parsing and syntax highlighting, provides code completions with icons, displays hover-based symbol definitions and docstrings (rendered with highlights and markdown), and includes a symbol outline sidebar for navigation. A floating peek widget enables jumping to symbol definitions or declarations, and symbol references are visible within the open editor. Source control is integrated with a split diff viewer for changes and staged changes, easy access to essential git operations (push, pull, commit, stage, unstage, branch switching), the ability to clone repositories and open them in the editor, and (currently disabled) support for viewing GitHub issues and pull requests. Search is powered by ripgrep for fast queries, accessible from the status bar, with support for regex, case-sensitive matching, and the ability to replace individual occurrences or all at once. Floating find-replace widgets enable searching within open editors. Debugging includes setting breakpoints across files, an inspection panel for runtime variables with the ability to modify them, call stack visualization and exception tracing, full DAP (Debug Adapter Protocol) client integration, and a built-in Python debugger; additional debuggers can be registered through extensions. The editor supports split markdown editing with plain HTML rendering, relative line numbering, vim mode, editor config support, and drag-and-drop for opening files and folders. Extensions are managed through a GUI, searchable within the editor, and can be bootstrapped via CLI commands and templates. A new extension marketplace website is available for discovering and sharing extensions. Biscuit is released under the MIT License and welcomes contributions; the documentation provides guides for developers and an API reference, with a contributing guide explaining the project structure and environment setup. Extension development is documented separately, and developers are encouraged to extend the editor with custom functionality.

Biscuit positions itself as a lightweight alternative to heavier IDEs and code editors by maintaining a sub-20 MB footprint while embedding AI agents at the core. The integration of multiple LLM providers—Gemini and Anthropic Claude variants—reflects the current landscape where developers want flexibility in which AI backend powers their tools. The agent system is grounded in practical file and codebase operations: reading, editing, searching, and executing terminal commands, rather than abstract capabilities. The extension marketplace and extensible architecture (documented for building custom extensions, registering language servers, and adding LLM providers) suggest a strategy to let the community shape the editor's capabilities without bloating the core. The project is MIT-licensed and actively soliciting contributions, which indicates an open-source development model aimed at building a developer community.

No comments yet. Be the first to share your thoughts!