Void: the open-source Cursor alternative

Void is a fork of Visual Studio Code that sends your code and prompts directly to the LLM provider you choose — OpenAI, Anthropic, Google, or a local model via Ollama — with no proprietary backend in between. It is free, MIT-licensed, and backed by Y Combinator. Evaluate it as a stable-but-frozen tool rather than a fast-moving one.

Void's editor UI — a VS Code fork with Agent Mode, Gather Mode, and bring-your-own-model chat.

Try Void →

Quick facts

What Void actually is (and what it isn't)

What it is. A drop-in-feeling VS Code replacement that adds AI features (autocomplete, inline edit, chat, agents) without routing your code through a company-owned server. You provide your own API key, or point Void at a local model running through Ollama, and everything stays on the network path you control.

What it isn't.

• Not a hosted service. There is no "Void cloud" to log into. No account, no telemetry endpoint for your prompts.
• Not a Cursor reskin. The signature Cursor UX (composer, tab suggestions, command-K) is replicated, but the wiring is different — Cursor routes through its own backend; Void does not.
• Not a VS Code extension. It's a full fork. You install it as a standalone app and import your VS Code settings in one click. (If you'd rather stay in VS Code, look at Cline or Roo Code instead.)

Void's distinctive features

These are the capabilities that separate Void from generic "open-source VS Code fork + AI" claims.

Agent Mode. Multi-step autonomous edits across your codebase. The agent reads files, writes changes, and runs through a plan without you approving each diff individually. Works with any connected model — including open-weight models that don't natively support tool calling, which is unusual for the category.

Gather Mode. A read-only variant of Agent Mode. The agent collects context from your codebase (files, symbols, dependencies) but cannot modify anything. Useful for exploratory questions about large unfamiliar repos where you don't want surprise edits.

Checkpoints for LLM changes. Every AI-driven modification is checkpointed. If a prompt goes wrong — a refactor that breaks three files, a rename that missed a callsite — you roll back to the pre-edit state without relying on Git stashes.

Fast Apply. Accelerated diff application for the "suggested edit" pattern. Instead of regenerating the full file, Void applies the model's change as a structured patch. Noticeable on edits in 1000-line files, where naive full-file regeneration is the main latency bottleneck in other forks.

FIM (fill-in-the-middle) model support. Native support for FIM-style completion models in Tab autocomplete, not just chat completion models. This is the correct API shape for autocomplete and most generic AI forks get it wrong.

Lint error detection. Integrates the editor's own diagnostics into the agent loop, so the model sees lint/type errors after it applies a change and can self-correct instead of leaving you with red squiggles.

One-click VS Code migration. Themes, keybindings, and settings transfer from an existing VS Code install in a single step. In practice this runs in under a minute on a typical customized settings.json.

Void vs Cursor vs Windsurf vs Zed — feature matrix

The one block a directory page can build that single-product pages structurally cannot.

Verify each row against current vendor documentation before making a buying decision. The active-development row is the one that shifts most frequently.

For a wider view of the category, see the full AI IDEs directory.

Strengths

• No vendor in the middle. Prompts and code travel directly from your editor to the provider endpoint you configure. If you've ever had to answer "where does our source code go?" in a security review, this is the shortest answer in the category.
• Any model, including offline. OpenAI, Anthropic, Google, xAI, OpenRouter, DeepSeek, Qwen, plus anything you can run locally through Ollama. If your laptop loses internet, the local-model path keeps working — Tabby and Continue are the only serious peers here.
• Familiar ergonomics. You keep your VS Code keybindings, themes, and settings. The learning curve is roughly "install, import, connect a model."

Weaknesses

• Development is paused. See the status banner. The repo is still public and buildable, but do not plan a roadmap around upstream fixes landing soon. If you need an actively-maintained alternative in the same shape, Windsurf and Zed are the closest equivalents.
• Ecosystem is narrower than Cursor's. Standard VS Code extensions install cleanly because of the shared format, but Void-specific integrations (MCP servers, third-party plugins explicitly targeting Void) are thinner on the ground.
• Community-only support. GitHub Issues and Discord. No commercial SLA. Acceptable for indie projects and research; likely disqualifying for regulated enterprise deployments that require a vendor contract.

When to choose Void (and when to skip)

Pricing

Void is 100% free. MIT-licensed on GitHub. No paid tier exists.

Your only running cost is whatever you pay the LLM provider you connect (OpenAI, Anthropic, Google, etc.) — or $0 if you run local models through Ollama.

Compare with Cursor's subscription tiers, Tabnine's paid plans, or Augment Code's $20+/month floor.

Install & first-run (5 minutes)

1. Download the Void build for your OS from voideditor.com.
2. Import VS Code settings when prompted — themes, keybindings, and settings.json move over in one click.
3. Connect a model. Pick one:
   • Hosted: paste an API key from OpenAI / Anthropic / Google / OpenRouter.
   • Local: install Ollama, ollama pull llama3.1 (or qwen3, deepseek-coder), and point Void at http://localhost:11434.
4. Open a project folder. Void indexes it automatically.
5. Try the three entry points:
   • Tab — accept autocomplete
   • Ctrl+K — inline edit on a selection
   • Ctrl+L — contextual chat with the file or repo

If something breaks at step 3, it is almost always a provider credential issue, not a Void issue. Verify the key works against the provider's own playground first.

Switching from Cursor to Void — what you keep, what you lose

Common Void setup mistakes

• Connecting to OpenRouter without a spend cap. OpenRouter will happily bill a runaway agent. Set a monthly limit in your OpenRouter dashboard before you attach it to Agent Mode.
• Running a 70B model on 16 GB RAM. Ollama will load it with heavy swap; you will see tens-of-seconds latency per keystroke. Start with a 7–8B model (Llama 3.1 8B, Qwen 3 7B) and scale up only if your hardware supports it.
• Pointing at a chat-completion endpoint for autocomplete. Tab autocomplete wants a FIM-capable model. Wire a chat model into chat, and a FIM model (e.g. DeepSeek-Coder via a FIM endpoint) into autocomplete — not the same model for both.
• Assuming "paused" means "broken." The last stable build runs fine today. Paused means no upstream features, not a broken editor. Decide based on whether you need forward progress, not on whether the current binary works.

FAQ

Is Void free? Yes. Void is MIT-licensed and always free. You only pay the LLM provider you connect — or $0 if you run local models via Ollama.

Is Void still being developed? Active development is paused as of YYYY-MM-DD (see the status banner). The repo remains public and buildable; no deprecation notice has been posted. The last stable release continues to work.

How is Void different from Cursor? Void has no backend of its own. Your code goes directly from the editor to the model provider you chose. Cursor routes prompts through its own backend, which is the architectural difference that drives most of the privacy, cost, and lock-in tradeoffs between them.

Can I use local models in Void? Yes, via Ollama. Any Ollama-runnable model works — Llama, Qwen, DeepSeek, Gemma. Fully offline. The closest peers for this are Tabby and Continue.

Will my VS Code settings transfer? Yes. Void offers one-click import of themes, keybindings, and settings.json. Most VS Code extensions install cleanly afterwards.

Should I pick Void or stay in VS Code with an extension like Cline? If you want the Cursor-style forked-editor UX (composer, tab model, integrated agents), choose Void. If you'd rather keep your existing VS Code install untouched, Cline and Roo Code deliver similar agent workflows as extensions.

What about MCP (Model Context Protocol) support? Basic MCP integration exists in the repo but is not as broad as Cursor's. Verify the specific MCP servers you need are supported before committing. If MCP coverage is a hard requirement, Cursor and Claude Code are ahead.

Verdict

Use Void if (a) data sovereignty is a hard constraint, (b) you can tolerate paused upstream development, and (c) you are comfortable wiring up your own model provider.

Choose Cursor, Windsurf, or Zed if you want an actively-maintained editor with polished, supported UX and don't mind a vendor backend.

Stay in VS Code with Cline or Roo Code if a full fork is more disruption than you want.

Sources

• voideditor.com — official site, feature list, pricing
• github.com/voideditor/void — repository, commit history, release cadence
• Y Combinator company page — company background, founders, funding
• InfoQ: The Void IDE, Open-Source Alternative to Cursor, Released in Beta — independent beta-release coverage
• Hacker News launch discussion — community reception and critique

Similar alternatives in AI IDEs

• Windsurf — closed-source, active, polished, the most direct Cursor competitor.
• Zed — Rust-built, from-scratch editor with integrated AI and multiplayer editing.
• Tabby — self-hosted, on-premises AI assistant for teams that can't rely on cloud endpoints.
• Continue — open-source assistant that plugs AI into your existing IDE instead of replacing it.
• PearAI — another open-source AI code editor, with zero data retention as its headline claim.
• Melty — open-source AI editor with deep Git integration.
• TRAE — actively-developed AI IDE from ByteDance.
• Opcode — desktop GUI that wraps Claude Code into an agentic editor.

Browse the full AI IDEs category or compare with IDE Extensions and CLI Agents. More context on picking between these categories is in the blog.

Last verified: 2026-04-22. This page is updated quarterly. Found an error or want to contribute? Submit an update.