rtrvr.ai v2: FULLY FREE Browser Use AI Agent! Automate ANYTHING! (Ranked #1 AI Web Agent)

Key Concepts

• Retriever: An AI web agent that automates tasks within a browser or the cloud.
• MCP (Multi-Cloud Platform): A framework enabling cloudscaling and multi-agent intelligence.
• Cloudscaling: The ability to scale operations across multiple cloud environments.
• Multi-agent Intelligence: The coordination and collaboration of multiple AI agents.
• Local-First Architecture: Running processes primarily on the user's local machine for privacy and security.
• DOM-based Intelligence: AI's ability to understand and interact with the Document Object Model (DOM) of web pages for accurate interactions.
• Web Bench: A benchmark used to evaluate the performance of AI web agents.
• Slash Commands: Commands that can be used within the Retriever interface to trigger specific actions.
• Record and Repeat Task: A feature allowing users to record a sequence of actions and have the AI repeat them.
• WhatsApp Bot: A feature allowing users to trigger automations and fetch data via WhatsApp.
• Chrome Extension: A browser add-on for Retriever, enabling remote browser control.
• Cloud and API Platform: A web-based platform for scaling automations to the cloud.
• Agentic Tabs: Tabs within the browser that are actively being controlled or utilized by the AI agent.

Retriever Version 2: The Next Evolution of AI Web Agents

This video introduces Retriever version 2, a significant upgrade to the AI web agent that transforms how users interact with the web. Previously known as a powerful local-first browser extension, Retriever now leverages MCP, cloudscaling, and multi-agent intelligence to offer enhanced capabilities.

Enhanced Browser Control and Remote Execution

The core innovation in Retriever version 2 is that the user's browser itself becomes a remote MCP server. This allows other applications and AI agents to directly control the browser remotely.

• Example: A user can instruct Retriever to send a connection request to Google CTO Jeff Dean. The agent can then autonomously open LinkedIn in the background, locate Jeff Dean's profile, and send the request using securely stored credentials.
• Key Advantage: This local-first architecture allows Retriever to handle websites that typically block cloud bots, such as LinkedIn, while simultaneously executing multiple tasks across different browser tabs.
• Scalability: Tasks can be saved as shortcuts for instant reruns or scaled to the cloud with a single click.

Performance and Benchmarking

Retriever distinguishes itself through its impressive performance metrics on various benchmarks.

• Web Bench Success Rate: 81.39% overall success rate.
• Task Completion Speed: A rapid 0.9 minutes per task.
• Cost Efficiency: An ultra-low cost of 12 cents per task.
• Capabilities: Retriever excels in both reading and writing tasks, outperforming competitors like OpenAI and Anthropic.
• Technical Advantages:
  • Local-first architecture: Avoids bot detection.
  • Reuses authenticated sessions: Streamlines login processes.
  • Handles CAPTCHAs: Overcomes common web security measures.
  • Supports multi-tab workflows: Enables complex, parallel operations.
  • DOM-based intelligence: Ensures accurate interactions with intricate web page structures.

Accessing and Using Retriever

Users have three primary ways to access Retriever's functionalities:

1. WhatsApp Bot:

   • Functionality: Trigger automations and fetch data directly from a mobile phone via text messages.
   • Simplicity: Operates through straightforward texting.

2. Chrome Extension:

   • Functionality: Acts as a remote MCP server, allowing AI agents to control the browser remotely.
   • Installation: Easily added to Chrome from the Chrome Store with a single click.
   • Account Creation: Free to access, with options to create an account or sign up using Google.
   • API Access: Free API access is available via Gemini.
   • Core Features: Automating web tasks, extracting data, automating data entry, crawling pages, creating documents, calling APIs and MCPs, and working with spreadsheets.
   • Natural Language Prompts: Users can automate tasks by providing natural language instructions.
   • Contextualization: The agent can be provided with context from specific web pages (e.g., a YouTube channel page and a newsletter) to perform tasks asynchronously across them.
   • "Teach a Trick" Feature: Allows users to record new tasks for the AI to learn and perform.
   • MCP URL: A unique URL generated by Retriever that connects the browser as a tool to other AI agents or apps for remote control.

3. Cloud and API Platform:

   • Functionality: Scales local automations to thousands of browsers in the cloud.
   • Applications: Extracting website data at scale, processing spreadsheets at scale, and automating any workflow at scale.
   • Execution: Tasks can be described in natural language and executed rapidly using different models (e.g., Flashlight, Flash, Pro).
   • Configuration: Allows setting default proxies and integrating tools similar to the extension.
   • Output: Extracted data can be opened directly in Google Sheets.
   • Hybrid Approach: Offers a flexible combination of local extension and cloud platform usage.

Demonstrations and Use Cases

The video showcases several practical applications of Retriever:

• Scraping YouTube Channel and Newsletter Content:

  • Prompt: "Scrape the contents of my YouTube channel page as well as my newsletter. Scrape the data as markdown."
  • Process: Retriever analyzes the request, plans execution, and opens agentic tabs for the YouTube channel and newsletter links.
  • Result: Rapid extraction of YouTube links, total views, and upload dates, along with a summary of newsletter posts. The task can be shared, copied, or saved as a shortcut.

• Extracting AI Startup Data from Angel.co:

  • Prompt: "Extract contents off of angel.co, finding different sorts of AI startups founded after 2023. Have it precisely execute this task that is very detailed, scrolled until all the results are loaded, and then extract each company name, description, location, and a couple of other fields."
  • Process: The task is executed at scale across thousands of cloud browsers.
  • Result: Comprehensive data extraction, including parent website, source website, location, tagline, industries, and other requested fields, presented in a Google Sheet.

Conclusion and Recommendations

Retriever version 2 represents a significant advancement in AI web automation, making complex tasks more accessible and efficient. The tool offers a powerful combination of local privacy and cloud scalability.

• Recommendation: The presenter highly recommends Retriever as one of the best AI web agents available and encourages viewers to explore it using the provided links.
• Call to Action: Viewers are encouraged to subscribe to the World of AI newsletter, join the Discord server, follow on Twitter, like the video, and check out previous content.