MiroFish
Multi-agent simulation engine: you feed it a seed text (launch page, pricing, draft policy, news item), it builds a world of thousands of LLM agents each with a personality and a memory, runs it, then writes a prediction report. AGPL-3.0 licence, about 76.3k stars on October 5, 2026.
Strengths
- Makes a launch face varied profiles before the first real customer: who buys, who bounces, and the reason each agent gives
- You can inject a variable (another price, another hook) mid-simulation and compare the reactions
- Open source, local interface or Docker: nothing forces you to hand your launch to a third-party service
Limitations
- Does not replace real customers: an LLM simulation has no statistical validity, the video says so itself
- You need a model API key (the repo recommends Qwen-plus) and a Zep Cloud account for agent memory, and simulations burn a lot of tokens
- AGPL-3.0 licence: if you expose it as a network service you must publish your modifications. Read it before any commercial use
- The repo still announces "coming soon" examples: a young tool (created November 2025), to treat as a lab
Best for
- PMs and founders with a launch page or a price grid who want to find the objections before paying for them for real