VS Code fork · AGPL-3.0 source

AI inside the editor

No subscription and no telemetryYour keys and your models, local ones includedEverything the agent sees and does stays under your control

Download for WindowsInstaller .exe · v1.29.1 · 162 MB Download for macOSApple Silicon, .dmg · v1.29.1 · 299 MB

or with one commandwinget install VibeBrains.VibeIDE

No accountExtensions from Open VSXWindows and macOS

Version v1.29.1 of October 28, 2026

$0 a month

You pay only your model provider, for the tokens you use

0 bytes

No telemetry, keys live in the OS keychain

20 providers

Built in out of the box, your own one is a single file away

Any model

A new model todayNot in the next editor release

The .vibe/providers.json file adds, overrides or turns off a provider and its modelsThey work on par with the built-in ones, and the model list comes live from the APINo key — run models locally with Ollama, LM Studio or vLLM

Example: your own model gateway in one file

{
  "version": 1,
  "providers": [{
    "id": "my-gateway",
    "name": "LLM Gateway",
    "protocol": "openai",
    "baseURL": "https://llm.company.local/v1",
    "apiKeyEnv": "GATEWAY_KEY"
  }]
}

Comparison

Why VibeIDE and not Cursor, Windsurf or Copilot

The same jobs done with no monthly feeAnd without sending your code to the editor vendor

Criterion	VibeIDE	Cursor	Windsurf	Copilot
Open source	AGPL-3.0	closed	closed	closed
Subscription	not needed	$20 a month	$15 a month	$10 a month
Your own API key	any provider	limited	limited	no
Custom providers without a rebuild	`.vibe/providers.json`	base URL only	no	no
Local models	built in	via an extension	no	no
Telemetry to the vendor	none	yes	yes	yes
Multi-agent work	built in	basic	no	no
Extension marketplace	Open VSX, no account	VS Marketplace	VS Marketplace	VS Marketplace

Features

What it does

Not a list of checkmarksBut the reasons this editor was written

An agent with autopilot

Takes a task to the end on its own: multi-agent orchestration with locksPlans survive an IDE restart, and on a limit the team saves its work and carries on

Token savings

Terminal and MCP output is compressed before it reaches the context, errors and results are kept word for wordYou see how many tokens were saved

Offline voice input

Dictation into chat, editor and terminal, recognised right on your computerRussian and English, the audio goes nowhere

Preview without a deploy

Static files, a framework dev server or a Docker setup right in the IDEClick an element of the page and a draft of the edit is already in the chat

A spec before the code

A big feature is described by a spec first: what we build and howDuring the build, edits outside the declared scope are paused, so the agent does not spread over the project

Videos and calls in the chat

The /watch command downloads a video, cuts frames at scene changes and transcribes the audio locallyThe model sees what was on screen instead of retelling subtitles

Private by default

No Microsoft telemetry, and secret detection runs before anything goes to a modelA session token limit and a context gauge, so the bill is never a surprise

Compatible with VS Code

Your habits, shortcuts and extensions stay with youOpen VSX without a Microsoft account and an IntelliJ-style keymap for those coming from JetBrains

Autopilot

You set the taskThe agent takes it to the end

Plan, edits, tests and review, without a nudge at every stepA model limit does not wipe the work: progress is saved and the agent carries on where it stopped

The family

Three toolsOne set of rules

Both IDEs read the same project rules and settingsAnd the agents' memory links them across sessions and machines

You are here

VibeIDE

An editor built on VS Code with its own agent

VibeIDEAAn IDE on the IntelliJ platform for PHP and TypeScriptOpen VibeMemoryShared memory for agents: context outlives sessions and machinesOpen

Install

Up and running in a minute

Download and run
Windows — the .exe installer or the winget commandmacOS — a .dmg image for Apple Silicon
Connect a model
A key for Anthropic, OpenAI, Gemini, DeepSeek or any OpenAI-compatible serverOr no key at all, with local models
Open a project and ask
The agent sees the codebase, edits files and runs the testsAnd at any moment you see what it is about to do

Latest builds

v1.29.1 — October 28, 2026macOS · DMG299 MB macOS · ZIP316 MB Windows · portable267 MB Windows · portable277 MB Windows · installer162 MB Windows · installer172 MB
v1.28.0 — October 22, 2026macOS · DMG299 MB macOS · ZIP316 MB Windows · portable330 MB Windows · installer200 MB
v1.27.0 — October 15, 2026macOS · DMG299 MB macOS · ZIP316 MB Windows · portable330 MB Windows · installer199 MB

All versions and release notes — on GitHub

First launch: the system will ask you to confirm

The installer is not signed with a publisher certificate: it is paid and tied to a companyThe system says it cannot verify the publisher, not that it found anything badYou confirm it once

Windows

In the SmartScreen window click «More info» — at first only «Don't run» is visible
Click «Run anyway»

Or install from the Microsoft catalogue, with no warnings

winget install VibeBrains.VibeIDE

macOS 15 Sequoia and newer

Drag the app to Applications and launch it, then click «Done» in the «Not Opened» windowWithout this attempt the button you need will not appear
Open «System Settings → Privacy & Security» and click «Open Anyway»
Confirm with Touch ID or your password and launch VibeIDE again — an «Open» button appears

Control-click → «Open» no longer helps in these versions: Apple removed this bypass starting with Sequoia

The terminal way and checksum verification — in the first launch guide

Honest about platforms

Windows and macOS on Apple Silicon are built and published with every releaseLinux builds from source for now, ready packages are in progress

Follow the issues on GitHub

Take the whole editor

Open source, no subscription, the vendor never sees your codeOne installer, and then it is just you, your project and your models