The reference shelf
Everything on this shelf earned its place: the product itself, its source and documentation, the recorded demos, and the research that explains why simulating a crowd of AI agents works at all. One or two sentences on each, so you know why you're clicking.
The product
- MiroFish — main site
- Where every scenario in this library actually runs. Ask a question, attach seeds, get a knowledge graph, a multi-agent simulation, and a prediction report. Start here if you start anywhere.
- MiroFish product overview (mirofish.ink)
- A single-page walkthrough of the product: use cases, the five-step workflow, and what a report contains. Useful to send to a colleague who wants the tour before the tool.
Source & documentation
- MiroFish on GitHub
- The full open-source codebase — graph construction, agent simulation, and report synthesis. Also the right place to file issues or propose a scenario for this library.
- README — setup & architecture
- The project's own documentation: what the system does, how the pieces fit, and how to run it yourself rather than using the hosted product.
Recorded demos
- Demo 1 — a complete run
- Question in, report out: the whole pipeline in one sitting. Watch this before your first run so the stages don't surprise you. (Also embedded on our homepage.)
- Demo 2 — interrogating a finished simulation
- The underrated half of the product: the follow-up chat. Shows how to push on a finished run with new questions instead of accepting the report as the last word.
The research behind it
- Generative Agents: Interactive Simulacra of Human Behavior (arXiv, 2023)
- The paper that made LLM-agent societies a serious research object — 25 agents in a simulated town producing believable emergent social behavior. The intellectual ancestor of tools like MiroFish.
- Social Simulacra (arXiv, 2022)
- The precursor idea: prototyping a social system by generating the crowd that would use it, so designers meet trolls and pile-ons before real users do. Directly relevant to why you'd rehearse an announcement.
- Agent-based modeling (Wikipedia)
- The decades-old discipline this all sits on: simulate individuals with simple rules, study the collective behavior that emerges. Good grounding for reading any prediction report critically.
- Multi-agent systems (Wikipedia)
- The broader computer-science frame: many autonomous agents, local decisions, global effects. Explains the vocabulary MiroFish's reports and this library both lean on.
Shelf policy: nothing gets listed that we wouldn't genuinely send to a colleague. When you've done the reading, run something at mirofish.us.