About this video

• Video Title: This AI Agent Just Did What I Thought Impossible | It LITERALLY Does Anything!
• Channel: Astro K Joseph
• Speakers: Astro K Joseph
• Duration: 00:15:32

Overview

This video introduces and demonstrates Retriever AI, an advanced AI agent capable of performing a wide range of tasks. The presenter highlights its ability to automate complex workflows, navigate websites, fill out forms, run code, scrape data, and integrate with other platforms like Cloud and WhatsApp. The video showcases various real-world use cases to illustrate the tool's power and versatility.

Key takeaways

• Retriever AI Capabilities: The AI agent can handle tasks such as filling out forms, navigating websites, running code, scraping data, and executing multi-step workflows using plain language commands.
• Accessibility: Retriever AI is accessible via a Chrome extension, a cloud interface, and even WhatsApp, offering flexibility in how users interact with it.
• Automation of Repetitive Tasks: The tool acts like a digital employee, automating mundane and repetitive tasks within a browser.
• Real-world Use Cases: The video demonstrates practical applications, including data extraction into Google Sheets, posting content on LinkedIn, running JavaScript code, comparing products on Amazon, and fetching stock market data.
• Integration and Customization: Retriever AI supports integration with MCP clients like Cloud, allowing for extended functionality. It also allows users to create shortcuts for frequent tasks and teach the AI new tricks.

You can access and use Retriever AI through three primary methods:

1. Chrome Extension: A browser extension that integrates directly into your browsing experience.
2. Cloud Version: A web-based interface accessible through their website.
3. WhatsApp: Integration allowing you to interact with the AI via WhatsApp messages.

Certainly. The video provides a detailed example of a complex multi-step workflow:

The prompt given to Retriever AI was: "Go to ycompinator.com/compies and filter for summer 2025 batch and keep scrolling till all companies are loaded and then extract two sheets for each company listed the name location tagline industries and URL."

This workflow involves Retriever AI performing the following actions:

• Navigating to a specific website (ycompinator.com/compies).
• Applying a filter (summer 2025 batch).
• Scrolling the page until all entries are loaded.
• Extracting specific data points (name, location, tagline, industries, URL) for each company.
• Creating a new Google Sheet.
• Populating that Google Sheet with the extracted information.

This task automates what would typically take a person at least an hour to complete manually.

Retriever AI integrates with other tools, such as Cloud, through MCP (Master Control Program) support.

Here's how it works and the functionalities it enables:

1. Adding Retriever as a Connector: You can add Retriever AI as a connector within an MCP client like Cloud. This involves copying an MCP URL from Retriever and adding it as a custom connector in Cloud.
2. Interacting via the MCP Client: Once connected, you can interact with Retriever AI directly through the interface of the MCP client (e.g., Cloud).
3. Behind-the-Scenes Execution: When you give a prompt in the MCP client, Cloud (or the respective MCP client) uses Retriever AI behind the scenes to perform the requested actions.
4. Extended Functionality: This integration allows you to leverage Retriever's capabilities within the environment of another tool. For example, the video shows a demonstration where Cloud uses Retriever's "execute JavaScript" tool to run a piece of code. This means you can use Retriever's powerful automation and execution features, but initiate them from your preferred integrated platform.

In essence, the MCP integration allows Retriever AI to act as a powerful backend engine for other applications, extending their capabilities by providing advanced AI-driven automation and task execution.

The "teach a trick" feature in Retriever AI allows users to train the AI agent to perform specific, potentially complex workflows that might not be covered by its existing capabilities or presets.

Here's how it works:

1. Initiating the Teaching Process: Users can click on a button labeled "teach a trick" within the Retriever interface.
2. Performing the Task Manually: After activating this feature, the user then manually performs the sequence of actions they want the AI to learn. This could involve navigating websites, clicking buttons, filling out forms, or any other series of steps.
3. AI Observation and Learning: Retriever AI observes these manual actions.
4. Saving the Trick: Once the user completes the task sequence, they can save it as a learned "trick" or command.
5. Autonomous Execution: In the future, the user can then ask Retriever AI to perform this learned trick, and the AI will execute the steps autonomously, just as if the user had performed them manually.

This feature essentially allows users to create custom automation routines for repetitive or intricate tasks, enhancing the AI's ability to handle personalized workflows.