MarketMirror

MarketMirror: AI Market Research & Marketing Simulation

Try your ads, prices, and sales pitch with simulated buyers before spending on a real campaign.

English · 简体中文

Open source and built for AI agents. Interviewed LLM buyer personas scroll simulated ad feeds, argue in an OASIS social simulation, answer a survey and hear your sales pitch. Ask in plain words (*"test these three ads on young parents in India"*) from Claude Code, Codex, Gemini CLI, Cursor or any MCP client, and get a go / iterate / no-go report in one offline HTML dashboard. Self-hosted: only the model calls (to the provider you choose) and optional public research leave your computer.

Built on MiroFish and CAMEL-AI's OASIS · listed in the official MCP Registry as io.github.DarkWorldCoder/market-mirror

[Open the live demo](https://darkworldcoder.github.io/market-mirror/demo/): the full dashboard from a 2,999-buyer simulated run, in your browser, no install. (Fitloop is an invented app; every number is simulated.)

Try it (Linux, macOS or WSL; needs Git, uv and Python 3.12):

git clone https://github.com/DarkWorldCoder/market-mirror.git && cd market-mirror

bash install.sh --configure # installs the engine, then asks which model provider to use ./mm start && ./mm cli quickstart --out scenarios/acme # 9 questions; prepares a small simulated market, then stops for review

The last line already makes model calls on the provider you chose, so start with a small market. If you chose a Claude Code or Codex login, run ./mm gateway start claude (or codex) before it. Already in an MCP client? Add the server with claude mcp add market-mirror -- uvx marketmirror (any client: command uvx, args ["marketmirror"]), then ask your agent to run doctor and setup. Full quick start.

If MarketMirror saves you one bad launch, a ⭐ helps other people find it. To hear about new tests and releases, use Watch → Custom → Releases.

▶ Watch the 60-second launch film (with captions) · vertical 9:16 cut · made with HyperFrames from launch-video/

Meet the buyers · What you can test · Screenshots · How it works · Example results · Evidence and limits · Get started · Documentation

What is MarketMirror?

MarketMirror gives you a practice audience for your next marketing decision. Describe your product and the people you want to reach. It creates AI-generated people who can react to your ads, compare prices, ask questions, raise concerns, and respond to a sales pitch.

You get a report explaining what interested them, what put them off, what to change, and what to test with real customers next. You can also watch the simulation and explore the results in a dashboard.

For example: you are launching a fitness app. Should the ad lead with “15 minutes a day” or “less than one gym visit”? Should the free trial ask for a payment card? MarketMirror lets you rehearse those choices before a live campaign.

The buyers are simulated. Their responses are generated by AI, so use them to develop and compare ideas. Confirm promising findings with real customers and small live experiments.

Who is it for?

Meet the simulated buyers

A simulated buyer is an AI-generated character representing someone in your intended audience. The character has a background and practical reasons for accepting, questioning, or rejecting an offer.

You choose the groups you want to explore—for example, students, working parents, and young professionals in selected countries. MarketMirror creates different people within those groups and can ask them about their lives before showing them your marketing.

An illustrative buyer: a busy parent who wants short home workouts, uses a budget phone, and worries about surprise charges. They might like a low monthly price but reject a “free” trial that immediately asks for a payment card. Another buyer may care more about workout quality than price. These are invented examples, not findings from real customers.

In tests within the same run, buyers can remember earlier experiences. Someone who disliked an ad can carry that reaction into a later survey or pitch. Separate comparison scenarios keep their own histories.

Detailed characters help you explore different viewpoints. They do not make the generated answers representative of a real population.

What you can test

You can repeat a comparison to see whether its result changes between runs. You can also use different AI providers and inspect where they disagree. Both help reveal uncertainty within the simulation.

The process at a glance

1. Describe your product. Provide a brief, documents, or a website: what you sell, who it helps, what it costs, and what supports your claims. You can also bring linked public feedback or campaign exports.
2. Choose your audience. Describe the groups and locations you want to explore. MarketMirror creates the simulated people. Review their profiles befo