What is market-fish?

MarketFish is a simulation tool that tries to predict whether a product idea will succeed by building a small artificial market instead of just asking a single AI model for an opinion. It creates over 128 AI shopper agents, each with its own identity, budget, emotional traits, and biases, and runs them through 30 rounds of shopping so their purchase decisions, drop-off patterns, and influence on each other can be observed like a miniature market. The project draws its design from six published research papers on agent-based simulation, including work on generative agents that remember and reflect on their own past decisions, a time engine that activates agents realistically instead of having everyone active every round, a recommendation system for personalized product suggestions, a cognitive loop with behavioral biases, a stress model that adjusts what agents are willing to pay under financial or social pressure, and a grounding layer that keeps agent decisions constrained by retrieved facts and rules. To use it, you clone the repository, copy the example environment file, add at least one API key for a supported large language model provider such as DeepSeek, OpenAI, or Anthropic, install the Python dependencies, and launch a Streamlit dashboard in the browser. From there you pick one of three modes: an exploration mode that lets the AI agents discover promising product directions from seed data on their own, a validation mode where you describe your own product idea and get back a survival score, buyer profiles, and a suggested price, and a hybrid mode that pits your idea against AI-generated competitor products in the same simulated market. There is also a command-line interface for running the same explore and validate workflows without the dashboard, including an option to reuse previously generated agents to cut down on API costs. The README compares MarketFish to a similar project called MiroFish, noting that MarketFish focuses specifically on product market prediction, runs as a single Streamlit app with no external memory service, and supports eleven different LLM providers across Chinese and global vendors. The project requires Python 3.12 or newer and is released under the MIT license, free for personal and commercial use.