GitHub radar
Rehearse the future in a digital world full of AI agents
MiroFish turns a real event into a parallel digital world of AI agents and hands you a forecast of how things will unfold.
An engine that builds a digital world out of thousands of AI agents — bots with their own memory and personality — and shows what happens inside it.
When it helps
Useful for rehearsing a decision — testing a policy or PR move in simulation before doing it for real — or just for playfully rewriting a story's ending. Skip it if you need a quick answer instead of a whole simulated world.
Pros
- Chat with any agent inside the simulated world afterward
- No need to host an AI model yourself — it runs through an API
- Works for serious forecasts and for playful what-ifs alike
Cons
- Needs a paid LLM provider key, plus a Zep key (Zep's free quota covers simple cases)
- Qwen-plus burns tokens fast — README suggests starting with simulations under 40 rounds
How to set it up — step by step
- 1Install Node.js 18+, Python 3.11–3.12, and uv — the Python package manager.
- 2Copy .env.example to .env and fill in an LLM provider key (README recommends Qwen-plus) and a Zep key.
- 3Run npm run setup:all so the agent installs every dependency — root, frontend, and backend.
- 4Run npm run dev and check the frontend loads at localhost:3000 and the backend at localhost:5001.
- 5Upload your seed material — a news story, policy draft, or story text — and describe in plain language what you want predicted.
- 6Read the forecast report and chat with any agent in the resulting digital world.
▌ More finds