OpenAI builds and deploys advanced AI models like GPT-4o for autonomous agents and workflows.
buttercup.sh
0Popularity
About buttercup.sh
buttercup.sh is a static HTML/CSS/JS web application that provides a sandboxed workspace for constructing and executing AI agents entirely within the browser. It offers a structured, week-by-week course format where each lesson introduces a single concept—such as tool-call loops, context compaction, or browser control—paired with executable code and hands-on exercises. Users can drag and drop text files into the workspace, define tools as plain JavaScript functions, and preview agent behavior in a sandboxed iframe. The tool supports multiple completion backends, including local models via Ollama, vLLM, or WebLLM, as well as cloud providers like Anthropic, OpenAI, or Google, with API keys stored in localStorage. Agents can perform tasks such as taking screenshots, navigating interfaces, and sending results back to the model, all while maintaining strict isolation between the workspace and the sandboxed preview. Sessions can be saved, undone, or exported as ZIP files, and context is automatically compacted to stay within model window limits. The interface includes commands like /help, /undo, and /new session for managing workflows.
Key features
- Weekly agent-building lessons with code and exercises
- Sandboxed iframe for agent preview and interaction
- Drag-and-drop file import for workspace
- Multiple backend providers (local and cloud)
- Context compaction to manage token limits
- Session management with save/undo/export
- Tool definitions as plain JavaScript functions
- Automatic model compatibility checks
Use cases
- Learning AI agent development through structured lessons
- Prototyping agent workflows with local models
- Testing agent interactions in a sandboxed environment
Pros
- No server required; runs entirely in-browser
- Supports both local and cloud-based AI models
- Sandboxed preview prevents external access to workspace
- Weekly structured lessons with executable code
- No signup required; uses localStorage for API keys
Cons
- Limited to text files in the workspace
- Local models require significant browser cache space
- No API access for programmatic control
- Single-language (JavaScript) tool definitions
Frequently asked questions about buttercup.sh
What is buttercup.sh?
buttercup.sh is a static web application that provides a sandboxed workspace for building and running AI agents entirely within the browser. It offers a structured, week-by-week course format where each lesson introduces a single concept paired with executable code and hands-on exercises.
Who is buttercup.sh designed for?
The tool is designed for anyone interested in learning how to build AI agents, from beginners to more experienced developers. It is particularly useful for those who want to experiment with agent development without requiring a backend or signup.
How does buttercup.sh handle API keys and model backends?
API keys are stored in the browser's localStorage, and model traffic is sent directly from the tab to the selected vendor. It supports multiple completion backends, including local models via Ollama, vLLM, or WebLLM, as well as cloud providers like Anthropic, OpenAI, or Google.
Can I use buttercup.sh offline?
Yes, once the model weights are downloaded into the browser's cache, the tool can work offline. Local models like those from Ollama, vLLM, or WebLLM are supported for offline use.
What kind of tasks can I perform with buttercup.sh?
Users can define tools as plain JavaScript functions, drag and drop text files into the workspace, and preview agent behavior in a sandboxed iframe. Agents can perform tasks such as taking screenshots, navigating interfaces, and sending results back to the model.
How do I get started with buttercup.sh?
To get started, open the tool in a browser, paste an API key (if using a cloud provider), and begin with the weekly lessons. The interface includes commands like /help, /undo, and /new session for managing workflows.
