The AI editor that bet on forking VS Code, then stopped

Void was an open-source, AI-native fork of VS Code. The pitch was simple: keep the editor you already know, swap the AI layer for one you point at any model or host, and send requests straight to the provider so your code never sits on a vendor’s backend. As of 2026-06 the project is archived. The README now opens with one line, “Void is now deprecated,” and the repository no longer accepts contributions. This page is about what Void tried, why that bet is harder than it looks, and what to reach for now.

What Void was

Void shipped a desktop app built from the full VS Code sourcecode. You could run AI agents over a codebase, checkpoint and visualize their changes, get fill-in-the-middle autocomplete, and bring any hosted model or local host into the loop. The differentiator was control: messages went directly to the provider you picked, with no Void-owned server retaining your data. For teams that could not send source to a hosted assistant, that was the entire appeal, and it landed. The project climbed to roughly 29,000 stars (as of 2026-06) from a standing start in late 2024.

Why it caught on, and what the curve shows

The star curve above tells a clean story. Void launched in September 2024 and crossed 5,000 stars within its first weeks, then surged again in May 2025, adding most of a second 10,000 in a matter of days. After that the line flattens: steady through late 2025, nearly horizontal into 2026 as work wound down toward the archive. Read it as a project that found its audience fast, rode the AI-editor wave of 2025, and then ran into the cost of what it had taken on.

The hard part was the fork, not the AI

The feature list never explains why a popular project stops. The repository’s own “Forking VS Code” notes do, if you read them as a maintenance ledger rather than a brag sheet:

  • VS Code’s build pipeline is private, so packaging, signing, and auto-update had to be rebuilt from scratch with custom GitHub Actions.
  • Mounting React and Tailwind inside the workbench is not possible in stock VS Code; Void had to extend the build to compile React and scope Tailwind by hand.
  • The AI provider code was written from scratch to support fill-in-the-middle autocomplete and custom responses, including grammars for <thinking> and tool tags, plus the IPC and CSP work to make it hold together.

Every one of those is a standing burden that Microsoft’s own release pace keeps reopening. The editor surface is the easy part; tracking a fast-moving upstream forever, while re-implementing a closed build and release pipeline, is the part that wears a small team down. Void’s deprecation is less a verdict on AI editors than on how expensive it is to maintain a serious VS Code fork.

What it leaves behind

Even archived, Void is one of the more complete open references for anyone forking VS Code, and the maintainers framed the wind-down that way: the code stays open precisely as a reference. Worth reading in the repo:

  • The build work to mount React and Tailwind inside the workbench.
  • A from-scratch AI provider layer with fill-in-the-middle autocomplete and grammars for thinking and tool tags, plus notes on IPC and satisfying CSP.
  • EditCodeService, which streams diffs into the editor token by token, and VoidModelService, which edits files in the background and syncs OS files with your text buffers.
  • GitHub Actions that package, sign, and auto-update a fork, normally hard because VS Code’s own pipeline is closed.

The repository keeps a VOID_CODEBASE_GUIDE and a HOW_TO_CONTRIBUTE for exactly this kind of reuse.

If you came here to use Void

You can still download old builds from the repository’s Releases page, but they are frozen and unsupported, and the issue tracker is closed. The maintainers point to a community Void Forks list for continuations. If what drew you was the underlying idea, an editor that keeps your code local and lets you choose the model, that niche is now served two ways: by actively maintained AI editors, and by pairing a standard editor with a local model runtime. A common setup is to run open models with Ollama and point an editor’s assistant at the local endpoint, which preserves the same “your code, your model” property Void was built around.

How it compares now

A like-for-like stats table is not useful here: Void is frozen, and its closest commercial peers are closed-source with no public repository to measure. The honest comparison is directional. Cursor and the hosted assistants won on polish and on never asking you to maintain anything, at the cost of routing code through their backends. Void chose the opposite trade, local-first and bring-your-own-model, and paid for it in maintenance. The category did not reject that trade; it moved it into editors and runtimes that did not also have to carry a VS Code fork. For the wider movement in this space, see what else is trending in LLM tooling.

FAQ

Is Void still maintained? No. As of 2026-06 the repository is archived and no longer accepts contributions, and the README states it is deprecated.

Can I still download Void? Yes, old builds remain on the Releases page, but they are frozen and unsupported.

What replaced Void? There is no official successor. The maintainers link a community Void Forks list, and the broader niche is covered by other AI editors and by pairing a local runtime like Ollama with a standard editor.

Why was Void deprecated? The README gives no single reason. The practical pressure on any VS Code fork is maintenance: tracking a fast-moving upstream while re-implementing a private build, signing, and auto-update pipeline is a heavy load for a small team.