Scrapybara

Scrapybara is a specialized platform designed to provide the essential infrastructure for building, deploying, and scaling AI-powered 'computer use agents'. These agents are advanced AI models, such as OpenAI's Computer-Using Agent (CUA), that can interact with a computer's graphical user interface (GUI) to perform tasks just as a human would—by moving a mouse, clicking buttons, and typing text. Scrapybara bridges the gap between these powerful AI models and real-world applications by offering reliable, on-demand virtual desktop environments where these agents can operate at scale.

The platform is engineered to eliminate the complexities of infrastructure management, allowing developers to focus solely on designing and refining their agent's logic and capabilities. By providing a unified interface and robust backend, Scrapybara empowers developers to automate a vast range of free-form computing tasks that were previously difficult or impossible to automate through traditional APIs.

How to use Scrapybara

Getting started with Scrapybara is straightforward for developers. The process typically involves these steps:

1. Installation: Begin by installing the Scrapybara SDK, which is available for both Python (`pip install scrapybara`) and TypeScript (`npm install scrapybara`).
2. Client Initialization: Initialize the Scrapybara client in your code. This client will be your main interface for interacting with the platform's services.
3. Launch an Instance: Use the client to instantly spin up a virtual desktop instance. You can choose from different environments, such as a feature-rich Ubuntu desktop, a lightweight Chromium browser, or a full Windows desktop. For example, `instance = client.start_ubuntu()`.
4. Execute Agent Actions: Use the `act()` function to send a prompt to your chosen AI model (like OpenAI CUA or Claude Sonnet). The function integrates the model with the virtual desktop instance, allowing it to perform actions based on your instructions. You can provide specific tools like `ComputerTool` or `BashTool` to define the agent's capabilities.
5. Monitor and Interact: The platform supports real-time interactive streaming, allowing you to monitor the agent's actions as they happen and even yield control if necessary. The results, including messages, steps taken, and final output, are returned in a structured format.
6. Playground Testing: For quick experimentation, Scrapybara offers a web-based playground where you can test different models and prompts on a live virtual instance without writing any code.

Core Features of Scrapybara

• Instant Desktop Instances: Instantly launch fully-featured virtual desktop environments, including Ubuntu, Windows, and a standalone Chromium browser, at massive scale.
• Speed, Scale, and Stability: The architecture is designed to start hundreds of instances in milliseconds and execute computer actions with minimal latency, ensuring high performance for demanding applications.
• Python & TypeScript SDKs: Provides easy-to-use SDKs for the most popular languages in AI development, simplifying integration and agent creation. A REST API is also available for custom integrations.
• Unified Agent Interface: A single, consistent interface for working with various computer use models, enabling powerful workflows with structured outputs and multi-turn conversations.
• Interactive Streaming & Control: Monitor agent activities in real-time and intervene when needed, offering full visibility and control over automated tasks.
• Session Persistence & Authenticated Access: Pause and resume instances to maintain state over long periods. The platform can also save and load authenticated sessions for websites, enabling complex workflows that require logins.
• First-Party Model Support: Close collaboration with AI labs like OpenAI ensures optimized, first-party support for state-of-the-art models like CUA.

Use Cases for Scrapybara

Scrapybara is ideal for any task that requires GUI-level automation. Examples include:

• Advanced Web Scraping: Go beyond simple data extraction to navigate complex, JavaScript-heavy websites, solve CAPTCHAs, and scrape data that requires user interaction.
• Automated Software Testing: Perform end-to-end UI testing by having an agent interact with your application just like a real user. The CodeCapy.ai project is a prime example, executing hundreds of parallel UI tests on Scrapybara instances.
• Robotic Process Automation (RPA): Automate complex business processes that involve legacy systems without APIs, data entry across multiple applications, or report generation.
• Deep Research and Data Collection: Deploy multiple agents in parallel to conduct deep research, gather information from various sources, and synthesize findings.
• General Digital Tasks: Automate everyday computer tasks like managing files, configuring settings on a web service (e.g., DNS records), or posting content to social media.

Advantages of Scrapybara

The primary advantage of Scrapybara is that it provides a robust, scalable, and reliable infrastructure layer for AI agents. This allows developers to avoid the significant engineering effort required to build and maintain such a system themselves. The platform's focus on speed and low latency is critical for creating responsive and effective agents. Furthermore, its unified API and support for leading AI models make it a future-proof choice for developers looking to leverage the latest advancements in AI-driven automation.

Pricing and Plans

Scrapybara operates on a freemium model, making it accessible for individual developers and scalable for large enterprises.

• Free Plan: Includes 10 compute hours, 100 agent credits, and allows up to 5 concurrent instances. Ideal for testing and small projects.
• Basic Plan: $29/month. Includes 100 compute hours/month, 500 agent credits/month, and up to 25 concurrent instances.
• Pro Plan: $99/month. Includes 500 compute hours/month, 2,500 agent credits/month, and up to 100 concurrent instances.
• Enterprise Plan: Custom pricing. Offers features like self-hosting, dedicated support, and a 99.9% uptime SLA for mission-critical applications.

An 'Agent Credit' is consumed for each step an agent takes (one assistant message with tool calls). Credits can be topped up at $0.04 each, or users can use their own provider's API key.