PREDICTYou commented this keyword
MiroFish
Upload a report or story, ask a what-if question, and MiroFish builds a simulated world of AI agents and returns a prediction report.
- Node.js 18+
- Python 3.11 or 3.12 and uv
- An LLM API key and a Zep Cloud API key
- Setup
- 30 min
- Level
- Advanced
- Stars
- 76.8k
What it does
MiroFish is a prediction engine built on many AI agents. You upload seed material, such as a news report, policy draft or a novel, and describe what you want to predict in plain language.
It builds a simulated world where thousands of agents with their own personalities and memories interact. You get a detailed prediction report and can chat with any agent in the simulation. There is a live demo you can try without installing anything.
Install it
- Install Node.js 18+, Python 3.11 or 3.12, and uv.
- Download the code from the GitHub repo and open a terminal in that folder.
- Copy the example settings file with: cp .env.example .env
- Edit .env and fill in LLM_API_KEY, LLM_BASE_URL, LLM_MODEL_NAME and ZEP_API_KEY. Any LLM API that uses the OpenAI SDK format works.
- Install everything and start the app with the commands below, then open http://localhost:3000.
Copy this
npm run setup:all npm run dev
Use it
- Upload your seed material, such as a data report or a story.
- Describe your prediction question in natural language.
- Read the prediction report, then chat with agents in the simulated world or with the ReportAgent.
Good to know
Simulations use a lot of LLM calls. The README suggests starting with fewer than 40 rounds. It recommends Alibaba's Qwen-plus model via the Bailian platform, and says Zep Cloud's free monthly quota is enough for simple use.
You can also run it with Docker: after creating .env, run docker compose up -d. The README does not state a license.
Want the next one first?
One new guide a week, straight to your inbox.
