Get started with VibeiDE

Install the desktop, connect your coding providers, and take one task from request to reviewed result.

Prerequisites

Before you start, install at least one command-line tool (CLI) on your computer: Codex, Claude Code or OpenCode. You only need one for your first task. Sign in to Codex or Claude Code, or configure at least one model provider in OpenCode, and check that it works in your terminal. A provider desktop app alone is not enough. Have a local project folder ready. Provider installation and sign-in are separate from the three-minute walkthrough.

Install and connect

Download the current installer from the public release page. On macOS, open the DMG and drag VibeiDE into Applications. On Windows, run the Setup installer or the portable build.
On first launch, VibeiDE reuses your installed tools and existing provider authentication. If Codex or Claude Code is missing or signed out, the desktop AI tool setup can help install it and sign in. OpenCode requires a separate installation and configured model provider.
Open a repository and set a modest per-provider concurrency limit. Existing provider subscriptions and model usage remain separate.

Queue a task you can verify

Choose Execute, a provider, model, and effort. Describe the intended result, relevant files, and checks to run. A new direct task starts a fresh conversation when its required capacity is free. Edit or reorder waiting requests from the project queue.

Inspect the run and review its result

Codex and Claude expose their original CLI interface in embedded terminals. OpenCode runs headless. Task Activity records reported work, and the global monitor shows active tasks across projects. Completed tasks collect in Ready for review.

Read the result and available patches, then run the relevant checks. Opening a result acknowledges completion, not code approval. Use supported conversation continuation for a related correction.

Choose a workflow

Execute and task queues for focused work.
Plan council for author, review, and revision before separate execution.
Turbo uses a fresh planning session followed by one fresh executor that owns the implementation graph.
Standup drafts a dated summary from successful project execution attempts.

Local storage and repository access

The orchestrator binds to loopback and stores tasks and artifacts on your machine. Provider CLIs retain authentication, while prompts and relevant context still go to the configured providers. AI sessions may have writable, unattended permissions. Use repositories and prompts you trust.

Project separation does not create Git worktrees or filesystem isolation. Sequence overlapping writes or prepare independent checkouts yourself.

Pricing and trial and referral terms.