Predivelli

Demo

About

The problem it solves Predivelli tackles a key problem in prediction markets: people often make decisions with limited information and personal bias. To fix this, it uses a team of AI agents—each focused on different areas—to analyze complex events from multiple perspectives. These agents then collaborate to reach a well-informed decision before placing a bet. For example, imagine there’s a major election coming up. Predivelli activates several agents: one reads news articles, another analyzes social media trends, a third looks at past election results, and another checks current betting odds on Polymarket. Each agent shares its findings, they debate the likely outcome, and then agree on the best prediction. Predivelli then places a bet automatically based on that group decision reducing bias and improving accuracy. But the project isn’t just a prediction tool. It’s also a hands-on guide for developers using Eliza OS. It shows how to build and manage intelligent agent teams, connect them to real blockchain platforms, and automate decision-making and trading. The project includes a custom Polymarket plugin, which will be shared in the official Eliza plugin directory.

How it's made

Challenges we ran into Building Predivelli came with several technical challenges, especially around integrating with Polymarket. One of the biggest obstacles was the absence of a dedicated and up-to-date Polymarket plugin. The existing GOAT Polymarket functionality was outdated and no longer suitable for this project. Because of this, I had to create a custom plugin that directly uses Polymarket’s CLOB API. This plugin made it possible to fetch live markets, place bets, and handle other important actions needed by the AI agents. Another major issue was dealing with the amount of data pulled from Polymarket. The data sets were too large for GPT-4.0 to handle in a single request, often exceeding its processing limits. To solve this, I had to filter and sort the data to focus only on the most recent and relevant markets, which allowed the agents to make timely and efficient decisions. Testing the CLOB API from the official Polymarket repositories and getting it to work inside the Eliza framework was also time consuming. There were bugs and unexpected behaviors that required a lot of trial and error to fix. Integrating everything smoothly while also building out the agent collaboration features made this part of the project particularly demanding. In terms of the development environment, I chose to work with the Eliza-develop-v1 version instead of the latest release. I found it to be more stable and better suited for building complex agent behaviors without unexpected breaking changes. Despite these challenges, the end result is a working Polymarket plugin that other Eliza developers can now use. It removes much of the friction I experienced and opens the door for others to build their own creative and intelligent agents that interact with prediction markets in meaningful ways.