Simulate public reaction to a price increase
Raise the flagship plan 20% and watch two simulated weeks of customer reaction before the real email goes out.
Read the card →MiroFish turns a question into a knowledge graph, a multi-agent simulation, and a prediction report you can interrogate. This club collects ready-to-run scenarios and resources for it — card catalog included.
Three well-thumbed cards from the drawer. Each one is a complete recipe: the setup, a copy-ready prompt, the seed files to attach, and the report signals worth reading first.
Raise the flagship plan 20% and watch two simulated weeks of customer reaction before the real email goes out.
Read the card →Pit your launch story against a simulated press, fan, and skeptic audience one week before the real one sees it.
Read the card →Retire the familiar face of the brand and find out which fan cluster defects — and which frame takes over — before it happens.
Read the card →You supply the question and, ideally, documents that name the actors and their incentives. That seed is the whole simulation's diet.
MiroFish reads the seed and wires the people, groups, and relationships it finds into a knowledge graph — the scenario's cast list.
Each graph node becomes an AI agent with its own persona. The agents react to your trigger and to each other, round after round.
The run is distilled into a prediction report — outcome, risks, narrative paths — that you can keep questioning in a follow-up chat.
The longer version, including what the follow-up chat can do, is on the about page. When you're ready, run any scenario at mirofish.us.
Embed blocked or JavaScript off? Watch the demo on Bilibili. A second recording, on interrogating a finished run, is listed under resources.
At mirofish.us — the main MiroFish site. Every page in this library is a recipe; mirofish.us is the kitchen. Paste the prompt, attach your seed files, and start the run there.
Keep the structure, swap the specifics. Each scenario names its audience groups, trigger, and time horizon — replace them with your own actors and dates, keep the questions at the end of the prompt, and attach seed documents about your situation rather than the template's.
Named actors with visible incentives. A page that says who the players are, what each one wants, and what they have said publicly gives the knowledge graph real material. Long background PDFs with no people in them add words, not accuracy.
Yes — MiroFish is an open-source project, and this library grows with it. Propose a scenario by opening an issue on the GitHub repository with your setup, prompt, and what the run should reveal.
On GitHub at github.com/666ghj/MiroFish. The repository contains the full simulation engine and the README documentation for running it yourself.