Void

Filed by the Vibe Code Weekly desk · Updated Nov 12, 2026

The story

Void is an open source AI IDE built on the VS Code codebase. Its bet is that developers should choose their own models, talk to them directly or run them locally, and keep their data out of a vendor's middle layer.

Void's site contrasts it with Cursor and Windsurf, saying Void does not relay messages through a private backend the way those editors do. The editor connects straight to whichever provider a developer picks, or to a model the developer hosts. For people who care where their code goes, or who want to swap models freely, that is the central reason to look at it.

On the editing side, Void covers the features people now expect from an AI IDE. Pressing Tab accepts an autocomplete suggestion. Quick Edit rewrites a highlighted selection in place. A chat panel offers three modes: Agent Mode, Gather Mode and a plain conversational mode. Because the editor descends from VS Code, existing themes, keybindings and settings can be imported with a single click, which lowers the cost of trying it alongside or instead of a current setup.

Recent additions focus on AI edits. Void offers checkpoints for changes made by an LLM, flags lint errors, supports fill-in-the-middle (FIM) models, and offers Fast Apply, which the project says works on 1,000-line files. The editor is currently offered as a beta download, with its code on GitHub and a Discord server for contributors and users.

Void says it sends requests straight to a model provider or a self-hosted model, not through a private backend like Cursor or Windsurf. For developers who care where their code goes, that is the main reason to look at it.

The good

Open source, with the code published on GitHub. Direct connections to model providers, with no vendor server sitting between the editor and the model. Works with self-hosted open models, which the project pitches as a way to avoid exhausting API credits. Familiar VS Code foundation with single-click import of themes, keybindings and settings. Checkpoints for LLM edits, lint error detection and Fast Apply for large files.

The catch

Still distributed as a beta. Model access is the user's job: either an account with a provider or a self-hosted model. As an editor, it leaves building, running and deploying the project to the developer's own workflow.

Who should read on

VS Code users who want AI assistance without rebuilding their editor configuration. Developers who want to connect directly to a model provider of their choice. Teams that prefer running open source models such as Llama, Qwen or DeepSeek on hardware they control. Engineers who want an AI editor whose source they can read and modify.

Who can skip it

People who want a hosted prompt-to-app service rather than a code editor. Buyers who need a stable, out-of-beta release.

Reader mail