Hey it's Paul, welcome to my 77th newsletter!

Are you fascinated by OpenAI's deep research capabilities but balking at the $200/month price tag?

Good news! This newsletter will walk you through setting up and running your own AI research agent, inspired by OpenAI's work, but completely free to use.

What This Agent Does (and why it's cool)

This AI agent acts like a dedicated research assistant. You give it a prompt (a question or topic), and it:

Searches the Web: It intelligently browses the internet, finding relevant information.
Iterates: It doesn't just stop at the first search. It analyzes the results and asks itself, "Do I need more information?" This creates a loop of searching and refining.
Generates a Report: Once it has gathered enough information, it compiles a complete report for you.
Runs on Replit: You don't need a fancy local setup. This runs in your web browser using Replit, a free online coding environment.

Step-by-Step Tutorial: Your Own Free Research Agent

This guide assumes you have a basic understanding of using a computer and navigating websites. You'll need a free Replit account and a Google AI Studio API key (also free).

Step 1: Fork the Replit Project

Duplicate the Replit Project on your own account (here's the link, 100% free).
Click the "Remix this app" button. This creates your own copy of the project that you can modify and run.

Step 2: Navigate to the Correct Directory

Once your forked Replit project is open, you'll see a console/terminal area (usually on the bottom or side).
Type the following command and press Enter:
```
cd web-ui
```
This command changes the directory (folder) to the one containing the web interface files.

Step 3: Run the Web Interface

Now, you need to start the web interface. Type the following command into the console and press Enter:
```
python webui.py --ip 0.0.0.0 --port 8080
```

Step 4: Access the Web UI

After running the command in Step 3, look for a "Webview" tab or section within your Replit interface.

Click on it. You should see a web page load – this is the interface for your AI agent! (You can maximize this window for a better view.)

Step 5: Configure the LLM (Get Your API Key)

In the Web UI, find the "LLM Configuration" section.
You'll need an API key to use a large language model. The tutorial uses Google's Gemini:
- Go to aistudio.google.com → Click "Get API Key" → Create a new API key → Copy this API key.
Back in the Replit Web UI:
- Select "Google" as your provider → Choose the "flash thinking exp model"
- Paste your API key into the appropriate field.

Step 7: Enter Your Research Prompt

Find the "Deep Research" tab in the Web UI.

Type your prompt into the text box. Be as clear and specific as possible. For example, the tutorial uses: "Complete report on reinforcement learning for training large language model."

Step 8: Set the Maximum Search Iterations

Also in the "Deep Research" tab, find "Max search iteration."
Set this to a reasonable number. The speaker suggests 15 iterations, which takes about 35-40 minutes. A higher number means more thorough research but a longer runtime.

Step 9: Run the Agent!

Click the "Run Deep Research" button.

You can also monitor progress in the "Shell" tab, which is the same place you enter commands.

Step 10: Wait and Review Your Report

The agent will now work independently. You can do other things while it runs.
The time it takes depends on the complexity of your prompt and the number of iterations.
Once finished, the report will be displayed (likely in the Web UI, but check the "Output" tab if you don't see it).

This method provides a fantastic, cost-effective way to leverage the power of AI for research. Enjoy exploring!

As always, if you have any questions or want to share your experiences with this AI Agent, feel free to reach out!

Have a great weekend and see you next week!

Paul

Promote your business in this newsletter

0 comments

AI are a paradigm shift (and the tools to build one)

This new AI chatbot is Private and Offline

Run a cutting-edge AI Model on your laptop

Turn Books (or files) into interactive Mindmaps with AI

Write the perfect prompt for reasoning AI models

See all posts from Paul Couvert