Not just extrapolation
Traditional forecasting often projects historical numbers forward. MiroFish is more useful when the next step depends on social reaction, coordination, incentives, beliefs, and timing.
MiroFish AI
MiroFish AI turns real-world seed information into multi-agent prediction simulations.
MiroFish is designed for questions where the answer depends on how people, groups, markets, communities, or fictional characters react over time. Instead of asking one model for a single forecast, it builds a small parallel world, populates it with agents, and studies what emerges.
How MiroFish AI works
- Seed intake: provide a prompt, PDF, Markdown file, news note, policy draft, product memo, market signal, or story outline.
- GraphRAG construction: MiroFish extracts entities, relationships, motives, constraints, and factual anchors from the seed material.
- Agent generation: the system creates independent agents with roles, memory, beliefs, and behavioral logic connected to the scenario.
- Simulation: agents interact across multiple rounds, producing narrative pressure, coalition shifts, objections, adoption curves, and second-order effects.
- Report synthesis: MiroFish turns the simulated trajectory into a readable prediction report with evidence, confidence signals, risks, and follow-up questions.
MiroFish AI vs ordinary forecasting tools
Not just a chatbot
A single LLM answer can sound confident but collapse many viewpoints into one voice. MiroFish keeps separate agents in play, which makes disagreement and emergence easier to inspect.
Useful for rehearsal
The output is not a guarantee. It is a rehearsal environment for decisions: compare scenarios, test assumptions, identify fragile groups, and ask better follow-up questions.
Start paths
Use the online MiroFish workspace when you want a managed browser experience. Use the GitHub repository when you want to inspect source code, Docker Compose setup, environment variables, license terms, or self-hosting constraints.