Actor groups
Check whether the report separates customers, critics, competitors, regulators, community members, creators, or investors instead of flattening everyone into one audience.
MiroFish how to use
Use MiroFish by turning one concrete scenario into a multi-agent simulation run.
A good MiroFish run starts with a clear question and enough seed context for agents to behave differently. The goal is not to ask for a generic prediction. The goal is to rehearse a decision, watch likely reactions, and inspect why the simulation moved that way.
First online run
- Write a decision-shaped prompt. Example: "If our SaaS product raises prices by 30% next quarter, how will customer sentiment, churn risk, and competitor narrative evolve?"
- Add seed context. Paste the relevant memo or attach a PDF, Markdown, or TXT file with product, audience, pricing, market, or policy context.
- Keep the time horizon explicit. Say whether you care about one week, one quarter, one year, or a multi-stage rollout.
- Run the simulation. MiroFish builds the scenario graph, creates agent groups, simulates interactions, and returns a structured report.
- Ask follow-ups. Drill into a stakeholder group, test a second variable, or compare the report against a different assumption.
What to look for in the report
Trajectory
Look for timing: immediate reaction, stabilization period, second-order effects, and delayed risks. MiroFish is strongest when the path matters as much as the final answer.
Actionable uncertainty
A useful report should name weak assumptions, missing evidence, and the variables that could flip the outcome. Treat those as your next test prompts.
Self-hosting path
If you prefer local control, inspect the MiroFish GitHub repository before deployment. Check Docker Compose, `.env.example`, model-provider settings, license terms, open issues, and the latest commit activity. The online workspace is better for quick tests; GitHub is better for source review and private infrastructure planning.