Rehearse your launch on test customers with MiroFish

Before you publish a landing page, a price or an ad, you can run it past a crowd of simulated customers and read how they react. MiroFish is the open-source tool that does it, and it had about 79,000 stars on GitHub on the day we read it.

MiroFish is not made by webridgeAI. Every detail below comes from its repository and README, read on 14 October 2026.

What it does

The README describes the whole loop in two lines. You upload seed material and describe what you want to predict in plain language. MiroFish returns a detailed prediction report and a simulated world you can interact with.

In between, it builds that world from your material: thousands of agents, each with its own personality, long-term memory and behaviour, who interact with one another. A ReportAgent then writes the report, and you can chat with any single agent or with the ReportAgent itself. The simulation engine underneath is OASIS, from the CAMEL-AI team.

You can see a finished run before installing anything: the project hosts a live demo of a simulation on a public-opinion event.

What you need

• Node.js 18 or newer
• Python 3.11 or 3.12
• uv, the Python package manager
• An LLM API key from any provider with an OpenAI-compatible API. The README recommends Alibaba's Qwen-plus model through the Bailian platform.
• A Zep Cloud API key. The README says the free monthly quota is enough for simple use.

Set it up

1. Clone the repository and open it:

git clone https://github.com/666ghj/MiroFish
cd MiroFish

2. Copy the example settings file:

cp .env.example .env

3. Open .env and fill in your keys. These are the four lines the README marks as required:

LLM_API_KEY=your_api_key
LLM_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
LLM_MODEL_NAME=qwen-plus
ZEP_API_KEY=your_zep_api_key

If you use a different provider, change the base URL and the model name to that provider's.

4. Install everything with one command:

npm run setup:all

5. Start it:

npm run dev

6. Open http://localhost:3000. The backend runs on port 5001.

With Docker instead: copy the settings file the same way, fill in your keys, then run docker compose up -d. It uses the same two ports.

Rehearse a launch

1. Upload what you are about to publish as the seed material: the text of your landing page, your pricing, or your ad copy.
2. Ask one plain question in the prediction box. For example: "How would early-stage founders react to this pricing, and who would pay for it?"
3. Keep the first run small (see below), then read the report.
4. Ask the ones who said no. Open a chat with two or three of the simulated customers who rejected it and ask them why.
5. Change one thing and run it again. A new headline or a new price, then compare the two reports.

The README's own examples use reports, news and stories as seed material. A landing page, a price or an ad fits what it allows (any uploaded material plus a goal in plain words), but it is not one of its published examples, so treat your first runs as an experiment.

Keep the first run cheap

• The README's warning, in its words: "High consumption, try simulations with fewer than 40 rounds first."
• Start with a short seed document and a narrow question. Every agent and every round costs LLM tokens.
• Use a cheaper OpenAI-compatible model for the first runs, and add rounds only once a small run tells you something useful.
• Set a spending limit with your LLM provider before you start.

What it is not

• Not your real customers. The crowd is simulated by a language model. Treat its reactions as a rehearsal that shows you questions to ask, not as data about the people who will see your launch.
• Not free to run. The software is open source, but every simulation runs on the LLM key you pay for.
• Not licensed for every use. MiroFish is released under AGPL-3.0. If you want to build it into a product or service for others, read the licence first.

Sources

• 666ghj, MiroFish, the repository, its README and its .env.example, read on 14 October 2026.
• MiroFish live demo, linked from the README.
• CAMEL-AI, OASIS, the simulation engine MiroFish is built on.