What MiroFish does
MiroFish is an open-source multi-agent simulation engine, incubated with support from Shanda Group. You upload source material such as a news report, policy draft or novel, describe what you want to predict, and it builds a knowledge graph, creates many AI agents with their own personas and memory, and lets them interact on simulated social platforms. It then writes a prediction report, and you can question the report agent or chat with individual agents. You run it yourself with your own LLM API key.
Best for
- Researchers testing how public opinion might react to an event
- Teams rehearsing reactions to a policy, launch or PR statement
- Writers exploring alternate story endings from a novel
- Developers who want a self-hosted agent simulation stack
Key features
Seed material to knowledge graph
It pulls entities and relationships out of your uploaded documents and builds a graph with memory using GraphRAG and Zep.
Generated agent personas
Each entity becomes an agent with its own profile, memory and behavior settings before the simulation starts.
Dual-platform social simulation
Agents interact in parallel on two simulated social platforms over many rounds, built on the CAMEL-AI OASIS engine.
Prediction report
A ReportAgent reviews the finished simulation and writes a report on how the scenario is likely to unfold.
Talk to the simulated world
After a run you can chat with any single agent or ask the ReportAgent follow-up questions.
Browser-based live demo
A hosted demo shows a prepared public-opinion simulation so you can look at the output before installing anything.
MiroFish pricing
MiroFish has a free plan. The full source code is free to download and run yourself under the AGPL-3.0 license. There is no paid plan from the maker on GitHub. Running simulations uses an LLM API you supply (any OpenAI-compatible API; the README suggests Alibaba's qwen-plus) and a Zep Cloud API key, which the README says has a free monthly quota enough for simple use. Prices checked on the official site on Nov 16, 2026.
| Self-hosted (open source) | $0 | AGPL-3.0; you pay for your own LLM API and Zep Cloud usage |
|---|
Source: MiroFish pricing page
Pros and cons
What it does well
- The code is open source, so you can run it on your own machine and inspect how predictions are made.
- It works with any LLM that offers an OpenAI-compatible API, so you are not tied to one model vendor.
- You get both a written report and the simulated world itself, so you can question individual agents.
Know before you sign up
- You need your own LLM API key and a Zep Cloud API key, and the README warns that simulations use a lot of tokens, suggesting runs under 40 rounds at first.
- Setup requires Node.js 18+, Python 3.11 or 3.12 and the uv package manager, or Docker.
- The AGPL-3.0 license requires sharing your source code if you offer a modified version as a network service.
Alternatives to MiroFish
LangChainBuild and manage AI agents.DifyBuild agent workflows on a shared canvas.
MetabaseAsk data questions and build verifiable dashboards.
WarpAutomate developer workflows with AI agents.
FicWeaveWrite multi-chapter AI fanfiction with story memory.
Domino Data LabBuild and operate enterprise AI models.
Questions about MiroFish
What is MiroFish?
MiroFish is an AI tool from github.com, listed in our Data & analytics category. Simulate how events unfold with many AI agents.
Is MiroFish free?
Yes, MiroFish has a free plan.
Where can I use MiroFish?
On Web.
What are some alternatives to MiroFish?
Tools that do a similar job include LangChain, Dify, Metabase and Warp.