AgentStrategy is a swarm of autonomous AI agents that analyze, trade, and promote a portfolio of AI agent tokens.
One agent is designed to analyze the top 100 AI agent tokens on Base. Every minute, it randomly selects one token and generates a detailed analysis report. The report covers social metrics such as Twitter mindshare momentum, engagement quality, and sentiment analysis. On the technical side, it evaluates price action, trading volume, and market structure. Each report also includes a buy, hold, or sell recommendation, along with a conviction level and suggested time horizon.
The trader agent is responsible for actively trading the top 100 AI agent tokens on Base. Every five minutes, it reviews insights from the analyst agent and determines whether to execute a trade. It factors in current holdings and market conditions to make the most strategic decision at the time.
The influencer agent monitors the latest reports and trade activity, crafting the most engaging tweet possible every hour. Over time, cookie data will be fed back into the LLMs to refine future tweets, optimizing for increased mindshare and engagement.
AgentStrategy will eventually accept LPs and aims to be the world's largest agent investment fund.
This project integrates several technologies to create a cohesive system for analyzing and trading AI agent tokens. It starts with the Cookie.fun API, which provides real-time data on AI agent tokens. In parallel, the system pulls relevant data from X (formerly Twitter) posts about each token to capture social sentiment and engagement metrics.
All data is processed and analyzed using OpenAI’s LLMs to generate a comprehensive investment analysis for one token every minute. The analysis is stored in Supabase, which serves as a centralized database for the agent network.
A second agent retrieves the analyst data from Supabase and leverages OpenAI to make trading decisions based on the latest insights. These decisions are executed automatically through Coinbase AgentKit, enabling seamless onchain trades.
Notably, combining social sentiment analysis with real-time trading decisions creates a self-reinforcing feedback loop. This dynamic system is still evolving, but the modular architecture ensures it can improve and scale over time.