Void

Void was the open answer to Cursor and for a while the right call if you wanted an editor talking to your own models with nothing in between. That is over. An editor that no longer ships updates is a security baseline frozen in time, and the daily trait that decides it now is the update badge that will never light up again; every week you keep it, the gap to upstream widens.

Pick Cline or Kilo Code inside stock VS Code and you get the same bring-your-own-model behavior with an editor that keeps patching, and Zed if you want the open-source editor with a team still behind it.

Every bug present when the repository was archived is now permanent, including the ones inherited from the VS Code it forked, because upstream fixes stop at the fork point. Combine that with an agent mode that has terminal execution and no sandbox and you have an unpatched binary that runs commands the model wrote, on your machine, forever.

Uninstall it before the next CVE in the upstream editor, which is a matter of weeks. What it did right: the model list was wide, Grok, DeepSeek, Qwen, Llama, Mistral and Gemma alongside the usual three, which is more range than most tools that are still alive.

Void made two documented decisions. 1. Fork VS Code rather than extend it, which buys deep integration at the cost of merging upstream forever; the cost is paid by whoever maintains the fork, and now nobody does. 2. Connect directly to hosted providers or local endpoints with no intermediary, which removes a vendor from the data path and a point of failure from the loop.

The second was correct and survives in every successor. The first is why an archived fork is a design without a maintainer for its most important dependency. The observation: the wrong decision was the structural one.

The case study for my least favorite pitch, open source as the marketing plan. Void had the product and the adoption, 28,813 stars, and no price on anything, so revenue was a future tense that never arrived. No acquirer buys an archived editor fork; the assets, the idea and the people, walked out without a term sheet, which is the exit that leaves nothing on the table because there was no table.

Likely path: the team resurfaces inside someone else's product, where the idea was always going to be a feature. Position: closed, and a reminder that stars are not a cap table.

Procurement is short. The repository was archived on 2026-06-02 and the README states the project is deprecated and no longer accepting contributions. That is an editor on sixty laptops that no vendor supports, no community maintains and no security team can patch, and the questionnaire ends there without reaching SSO, audit logs or retention, none of which it had.

Anyone who still has it installed moves to stock VS Code this sprint, and the endpoint tool confirms removal. Not yet, and in this case not ever.

Void did the architecture I keep asking for: the editor talked straight to my Ollama or vLLM server, no relay, no account, Apache-2.0, an MCP client, and every prompt in the source. The license lets me fork it, and the fork would work tomorrow exactly as it works today, which is the problem stated kindly.

Maintaining a VS Code fork means merging upstream every few weeks forever, and I am not signing up for that alone; the community that could share the load already left. Respect for the design; no fork from me.