Gemini 2.5 DEEP AGENTS: Powerful AI Agents BY Gemini Can DO ANYTHING For FREE!

Key Concepts

Gemini 2.5 Flash, AI Agents, Vector Shift, Pipelines, Nodes, Data Loaders, Data Transformation, AI Ops, Summarization, Categorization, Google Sheets Integration, Master Pipeline, List Operations, System Instructions, Deep Research Agent, Automation.

Gemini 2.5 Flash: A Powerful and Efficient Model

The Google DeepMind team has released Gemini 2.5 Flash, a highly efficient and fast AI model with built-in agent capabilities. It boasts a 1 million token context window, enabling it to process vast amounts of data. Benchmarks show it outperforms models like OpenAI's O4 Mini and Cloud 3.7 Sonnet across various tasks. Gemini 2.5 Flash offers native multimodal reasoning, tool use, and experimental features like image generation and text-to-speech. It integrates seamlessly with tools like Google Search and code execution. The model adapts its reasoning depth based on task complexity and developer-defined budgets, providing cost efficiency and intelligence.

Vector Shift: A Platform for Building AI Agents

Vector Shift is a platform used to create AI agents with a drag-and-drop builder. It's a node-based platform for building, deploying, and managing AI agents, applications, and chatbots. Users can sign up for free and access the main dashboard.

Creating a Deep Research Agent with Vector Shift and Gemini 2.5 Flash

The video demonstrates creating a deep research agent using Vector Shift and Gemini 2.5 Flash to scrape company websites and analyze the findings. The agent can be deployed within a Google Sheet to process leads, categorize target customers, and provide justification.

Step-by-Step Process:

1. Create a Pipeline: Start with a blank pipeline in Vector Shift.
2. Input Nodes: Place two input nodes: one for processing URLs (website links) and another for processing the description or content to classify the target customer. Name them "URL" and "Target Customer," respectively.
3. Data Loader (URL Node): Add a URL node from the "Data Loaders" category to scrape the URL from the input. Connect the "URL" input node to the URL node.
4. Data Transformation (AI Ops Summarizer): Place an AI Ops node configured as a summarizer to summarize the content scraped from the URL. Connect the URL node to the summarizer node.
5. Data Transformation (AI Ops Categorizer): Add another AI Ops node, this time configured as a categorizer, to identify if the company is a valid target audience. Connect the summarizer node to the categorizer node.
6. Categorizer Configuration: Provide context to the categorizer by stating "Categorize whether or not this company is my target customer or not." Insert a variable from the "Target Customer" input node to define the target audience. Define the classification categories as "Target Customer" and "Not the Target Customer."
7. Output Nodes: Place two output nodes: one for the target customer category and another for the justification. Rename them "Target Customer" and "Justification," respectively.
8. Connect Output Nodes: Connect both output nodes to the categorizer node. Configure the "Target Customer" output node to output the "Category" and the "Justification" output node to output the "Justification." Ensure "Include Justification" is selected in the categorizer node.
9. Rename and Deploy: Rename the categorizer node (e.g., "Target Customer Categorizer") and deploy the changes.

Creating a Master Pipeline with Google Sheets Integration

A master pipeline is created to call the target customer categorizer pipeline and process multiple leads from a Google Sheet.

Step-by-Step Process:

1. Create a Workbook: Create a new workbook in Vector Shift.
2. Google Sheets Node (Read): Place a Google Sheets node from the "Integrations" category to read data from a Google Sheet. Connect it to the Google account and select the sheet containing the website URLs and target customer information.
3. Pipeline Node: Place a pipeline node from the "General" category to call the "Target Customer Categorizer" pipeline. Enable "List Mode" to process multiple leads.
4. Connect Website URL: Connect the "Websites" column from the Google Sheets node to the URL input of the pipeline node.
5. Text Node (Target Audience): Place a text node from the "General" category to specify the target audience (e.g., "I sell into pharmaceuticals and I am looking for the target customer which is anyone that is in this industry").
6. List Operation Node (Duplicate): Place a list operation node from the "Data Transformation" category and configure it to duplicate the text from the text node. This is necessary because the data types don't match between the text node and the pipeline node.
7. Connect Target Audience: Connect the text node to the list operation node, and then connect the output of the list operation node to the "Target Customer" input of the pipeline node.
8. Gemini 2.5 Flash Integration: Within the "Target Customer Categorizer" pipeline, place a large language model node and select Gemini 2.5 Flash.
9. System Instructions: Provide system instructions to the Gemini 2.5 Flash model, such as "Explain why the company was categorized as a target customer, focus on relevant factors."
10. Connect LLM to Categorizer: Connect the large language model node to the categorizer node.
11. Google Sheets Node (Write): In the master pipeline, place another Google Sheets node from the "Integrations" category, this time configured as a column list writer, to write the results back to the Google Sheet. Connect it to the Google account and select the same sheet.
12. Confirm Selection: Confirm the selection to map the output fields (target customer and justification) to the appropriate columns in the Google Sheet.
13. Deploy and Run: Deploy the changes and run the master pipeline. The agent will process the leads, categorize the target customers, and provide justification, writing the results back to the Google Sheet.

Conclusion

The video demonstrates how to leverage the power of Gemini 2.5 Flash and the flexibility of Vector Shift to create a deep research agent for lead analysis and categorization. By combining the efficient AI model with a user-friendly platform, users can automate complex tasks and gain valuable insights from their data. The example showcases the tip of the iceberg, and users can further expand the capabilities of their agents with different nodes and integrations.