VibeKit

VibeKit is an open-source SDK designed to run coding agents like OpenAI Codex and Claude in secure sandboxes. It enables agents to write code, install packages, or open PRs safely, with built-in streaming, async tasks, and telemetry. MIT licensed and built with TypeScript, it offers zero lock-in. VibeKit allows developers to embed coding agents into their applications or workflows, supporting various sandbox providers such as E2B, Daytona, Modal, and Fly.io. It is ideal for creating internal tools, adding AI-powered features to apps, and streamlining prototyping workflows.

How to Use

To use VibeKit, developers can install the SDK via npm (npm i @vibe-kit/sdk). They then import VibeKit and VibeKitConfig to set up the agent type (e.g., 'codex') and other configurations. An instance of VibeKit is created, and methods like generateCode are called with a prompt and mode. Callbacks (onUpdate, onError) can be provided to handle streaming output and errors from the agent.

Key Features

• Run coding agents (OpenAI Codex, Claude) in secure sandboxes
• Open-source and MIT licensed with zero lock-in
• Supports streaming agent output to UI
• Customizable environment, allowing package installation
• Supports both code generation and asking questions
• GitHub integration for working on codebases and creating PRs
• Built-in streaming, async tasks, and telemetry
• Compatibility with various sandbox providers (E2B, Daytona, Modal, Fly.io)

Use Cases

• Internal Tools: Support debugging, refactoring, automation, onboarding helpers, CI bots, and internal dev tools.
• App Features: Add code generation or code Q&A to products, enabling users to scaffold apps, generate components, or explore code.
• Prototyping Workflows: Scaffold features, generate boilerplate, or explore different implementation paths for faster iteration.
• Integrations: Automate common setup tasks like adding analytics, wiring authentication, or configuring SDKs.