MiroFish How to Use: A Practical First Run

Best-fit use cases

• A developer is trying MiroFish for the first time and wants a safe workflow.
• A founder wants to simulate public reaction before a product launch.
• An analyst wants a repeatable graph-to-report process for a trading or forecast scenario.

Operating steps

1. Choose one scenario and write a narrow question with actors, timeline, and success criteria.
2. Build the graph from a small seed note and review entities before running agents.
3. Set the environment: API key, graph backend, memory store, and round limits.
4. Run a short simulation and inspect agent disagreements instead of chasing a single answer.
5. Generate a report, then ask one follow-up question that changes only one variable.

Common risks

• A huge seed file can create noisy entities and slow first-run debugging.
• Missing API or graph-store credentials can look like model failure when it is environment setup.
• Agent costs rise quickly when rounds, roles, and context windows are expanded together.