A group of engineers built a complete AI agent ecosystem using CrewAI and Google’s Gemini large language models inside a single notebook. The integrated system employs distinct agents configured for research, data analysis, content generation, and quality assurance. Each agent uses an AgentConfig object to set parameters like temperature, token limits, and custom toolkits. Users can customize these settings before any task runs.
The process begins with an automated pip install of key libraries such as crewai, google-gemini-client, openai, pandas, and rich. Each package is installed within the Colab runtime through a loop that confirms success before moving forward. Any installation error triggers an alert and retries the download.
Warning messages get suppressed to keep the log clean. The notebook imports the CrewAI core classes, including Agent, AgentConfig, TeamManager, and AgentGroup. It imports the GeminiClient wrapper from the google_gemini package along with utility modules like time, datetime, os, and json. These components will power the agent interactions.
A code block retrieves the Gemini API key from the Colab secure store. If no key is found, the script prompts the user for a secure paste. After storing the key in memory, the code executes a simple test call—sending a greeting prompt to verify authentication and confirm that the model is ready for work.
Next, two rapid-fire tasks run back to back. The first agent collects and summarizes the top five machine learning trends from recent publications. The second agent performs a detailed review of emerging technologies in sustainable energy, returning bullet-point findings. Each summary prints to the notebook so users can see real-time results.
The notebook then defines a minimal command-line loop. Users can enter commands such as research
Additional helper functions include download_results(task_id), which handles file path logic and metadata tags for local storage; print_markdown(task_id), which constructs titles, headings, and bullet subsets; and push_to_drive(task_id), which manages folder creation and overwrite checks on Google Drive. Each helper runs in a single line, making it easy to archive or share findings.
To showcase end-to-end automation, the notebook contains a demo script that iterates through three predefined topics and three task types. After each agent finishes, the output exports as Markdown. A brief pause between runs keeps the display organized and prevents rate limiting from the API.
What emerges is a reusable research framework that can launch multiple AI agents in series or parallel. Teams have the ability to run quick sketches, deep investigations, or live interactive sessions. Every result, from concise summaries and charts to raw JSON logs, can be fetched and stored with minimal code.
Developers can add custom agent templates for tasks like data visualization, sentiment analysis, or automated reporting. The modular codebase separates runtime logic from extensions. Official CrewAI and Gemini client docs provide detailed examples and configuration advice for advanced workflows.