MarketMirror: marketing experiments with simulated buyers, built on MiroFish/OASIS

[DarkWorldCoder]

Hi, I’m the maintainer of MarketMirror, an independent, open-source project built on MiroFish and OASIS. Thank you to both upstream teams for the simulation foundation. This is a community project, not an official MiroFish service.

MarketMirror adds a marketing workflow around that foundation:

• Describe a product and audience, then review the generated buyer profiles.
• Compare ads, prices, emails or sales pitches and inspect generated objections.
• Run the workflow through MCP tools, agent skills or a local CLI.
• Explore a saved HTML dashboard and use the findings to plan a small real customer experiment.

You can browse the saved example dashboard without installing anything. It shows an invented fitness app, Fitloop, with 2,999 AI-generated buyers. The page displays saved outputs; it does not run new simulations in your browser. Every buyer response and performance number in that example is simulated.

The source is AGPL-3.0 on GitHub, with MCP/plugin setup instructions. Actual runs require a configured model provider, and model usage can cost money. Synthetic responses do not establish demand, population representativeness or prediction accuracy.

I’d value feedback from people working with MiroFish/OASIS: which artifacts would you need to audit a comparison between two marketing messages—buyer profiles, prompts, seed/configuration, raw responses or something else? And what would make a first small run easier to inspect?

Codex helped prepare this post.