Void is an open-source AI code editor designed as a privacy-focused alternative to Cursor and Windsurf. It empowers developers to interact with AI models—such as Gemini 2.5, Claude 3.7, Llama, and DeepSeek—directly from their codebase without intermediaries, ensuring full data control. Built as a fork of VS Code, it preserves all existing themes, keybindings, and settings while adding advanced AI capabilities.
Technically, Void extends the VS Code codebase using TypeScript and integrates with LLM providers via direct API calls. It supports local hosting of open-source models, checkpointing of AI-generated changes, and agent-based automation with MCP tool access. Deployment options include downloadable binaries for desktop use and the ability to build custom forks using void-builder and the provided codebase guide.
What You Get
- Direct LLM Integration - Connect to any LLM provider (Gemini, Claude, Grok, Qwen, OpenAI) without routing through a private backend; all prompts and code are sent directly from your machine.
- Agent Mode - AI agent that can search, create, edit, and delete files and folders, with access to terminal and MCP tools for complex codebase operations.
- Gather Mode - Read-only AI mode that searches and analyzes code without modifying files, ideal for exploration and documentation.
- Checkpoint & Visualize Changes - Track and visualize AI-generated edits to your codebase with a visual diff interface, enabling rollback and review of AI-driven changes.
- Local LLM Hosting Support - Run open-source models like Llama, DeepSeek, Qwen, and Gemini locally without API credits or cloud dependencies.
- FIM Model Support - Built-in support for Fill-in-the-Middle (FIM) models for context-aware inline code generation and completion.
- VS Code Compatibility - Import all existing themes, keybindings, and settings from VS Code with a single click, ensuring seamless transition.
- Fast Apply on Large Files - AI suggestions and edits work efficiently even on files with 1000+ lines, optimized for performance in large codebases.
Common Use Cases
- Running a privacy-first AI-assisted development team - A startup uses Void to deploy AI coding assistants without exposing proprietary code to third-party servers, ensuring compliance with data governance policies.
- Developing with local LLMs for offline coding - A researcher runs Llama 3 or DeepSeek on a local GPU and uses Void to interact with the model directly for code generation without internet dependency.
- Building custom AI editors for enterprise use - An engineering team forks Void using void-builder to create a branded, internally hosted AI editor with custom tooling and model integrations.
- Experimenting with frontier models like Claude 3.7 or Grok 3 - A developer tests cutting-edge models directly in their IDE without waiting for official API access or paying for credits.
Under The Hood
Architecture
- Modular design rooted in Electron and VS Code’s extensible framework, with clear separation between main and renderer processes
- Layered architecture where product configuration drives behavior, abstracting remote services and extension endpoints through well-defined interfaces
- Dependency injection via service brands and decorators enables loose coupling between core systems and extensions
- Plugin-based extension model inherited from VS Code, supporting remote galleries and URI transformation pipelines
- Gulp-driven build system with targeted compilation phases for client, web, and CLI outputs, supporting complex asset pipelines
Tech Stack
- Hybrid architecture combining Rust for performance-critical backend components and TypeScript for frontend and tooling layers
- Built on VS Code’s codebase, leveraging Electron for desktop packaging, Monaco Editor as the core text component, and Playwright for browser testing
- npm-based dependency management with comprehensive license compliance tracking via cgmanifest.json and cglicenses.json
- Custom Node.js scripts and deemon orchestrate development workflows, while CodeQL and TSEC enforce static analysis and security across the monorepo
- Tight integration with Microsoft’s ecosystem, including VS Code extensions, marketplace integration, and platform-specific configurations
Code Quality
- Extensive test coverage with unit and integration tests for core UI and system behaviors, using reusable utilities for async and disposable patterns
- Strong type safety and modular design enforced through interfaces, events, and disposable patterns to minimize coupling
- Consistent, descriptive naming conventions and comprehensive code comments enhance readability and long-term maintainability
- Robust error handling with defensive programming, null checks, and graceful degradation across UI and async operations
- Strict linting, formatting standards, and copyright headers maintained across all code files
What Makes It Unique
- Native remote filesystem access via protocol-level abstractions, enabling seamless local-like interaction with remote resources
- Built-in secret storage with extension-scoped key management and sequenced operations to prevent credential race conditions
- Dynamic extension manifest parsing and NLS localization injected directly from DOM metadata, eliminating bundle overhead for built-ins
- Custom pixel-perfect color picker with global pointer monitoring and canvas rendering independent of external libraries
- Hierarchical test state engine with WeakMap-based caching for performance-critical large-node computations
- Early environment cleanup and UNC path support ensure consistent behavior across diverse deployment scenarios