MiroFish-Offline

Why we included this project

Drop a press release, policy draft, or earnings report into MiroFish-Offline and it generates hundreds of simulated social-media agents, each with a distinct personality, opinion bias, reaction speed, and memory of earlier events. Over simulated hours they trade posts and arguments while their opinions shift, and the engine tracks sentiment evolution, topic propagation, and influence dynamics in real time. A ReportAgent then interviews a focus group of the agents and summarizes how the narrative landed. Everything runs locally on your hardware, with knowledge graphs stored in Neo4j and language models served by Ollama, so there are no cloud API keys or per-token bills. Teams doing communications planning, product positioning, or market-sentiment exploration get a practical sandbox for stress-testing narratives before they go public.