The MiroFish Journal — notes on simulating how people react
MiroFish takes a question, builds a knowledge graph of the actors involved, runs them as AI agents in a multi-agent social simulation, and returns a prediction report you can keep interrogating. This journal explains each of those moving parts from first principles. The product itself lives at mirofish.us.
All articles
- July 10, 2026 · 7 min read
How MiroFish works: from seed material to prediction report
The flagship explainer: what actually happens between typing a question and reading a prediction — all five stages, no hand-waving.
- June 24, 2026 · 6 min read
Knowledge graph seeding: what MiroFish reads out of your documents
The graph is built from whatever you attach — here's what the extractor is looking for, and how to write seeds that give it something to find.
- June 5, 2026 · 7 min read
Multi-agent social simulation: why a crowd of AI agents beats one confident answer
The middle of the pipeline is where the interesting physics happens: agents reading agents, opinion cascades, and outcomes nobody scripted.
- May 20, 2026 · 6 min read
Reading a MiroFish prediction report: anatomy and use
The report is four sections and a temptation. Here's what each section is for, and how to use the whole thing without over-trusting it.
- May 2, 2026 · 6 min read
Simulation vs. a single chat answer: when do you actually need agents?
Not every question deserves a simulation. A field test for telling the two kinds apart — and why the difference is about loops, not difficulty.
- April 16, 2026 · 8 min read
Getting started with MiroFish: your first simulation, end to end
From blank page to first prediction report in one sitting — the question, the seed, the run, and what to do the moment the report opens.
Watch it run
This recording shows the follow-up chat against a finished simulation — watch it on Bilibili if the embed doesn't load. For the full pipeline from question to report, see the end-to-end demo.