MiroFish on Railway

Multi-agent simulation engine that turns documents into LLM agents, behind a login on Railway.

Official site · License: AGPL-3.0 · 2 Railway templates

Deploy MiroFish on RailwayCompare templates

MiroFish is an open-source multi-agent simulation engine. You give it documents and a scenario, it builds a knowledge graph, turns the people in it into LLM agents, runs them on simulated social platforms and writes a report. It builds on the OASIS simulation framework from CAMEL-AI.

People use it as a structured brainstorm: rehearsing how an announcement might be discussed, exploring audience reactions to a campaign, or running classroom and research experiments on opinion dynamics. The results are LLM-generated simulations, not forecasts.

MiroFish is licensed under AGPL-3.0. If you modify it and let others use it over a network, the licence asks you to offer them your modified source.

What MiroFish can do

  • Builds a knowledge graph from your documents
  • Turns entities into LLM agents with memory
  • Runs simulations on simulated social platforms
  • Generates a report from each simulation
  • Works with any OpenAI-compatible provider
  • Web interface with English and Chinese

What people use it for

  • Rehearsing how an announcement or policy draft might be discussed
  • Exploring audience reactions to a campaign using your own briefing documents
  • Classroom and research experiments on opinion dynamics
  • Playing out what-if scenarios from a story world

Videos

MiroFish Full Tutorial — Predict Any Scenario With AI · Tech With Tim
MiroFish: The First Open Source Swarm Intelligence Digital World Simulation Engine · DevsKingdom

Running MiroFish on Railway

The Railway template runs a Web service (Caddy) that serves the UI behind basic auth with a generated password, and the MiroFish backend on the private network with no public URL. A volume holds projects, simulations and the model cache. The template pins a MiroFish source commit and the v0.1.2 Python environment. You supply an LLM API key and a Zep Cloud key. Simulation cost is mostly your LLM and Zep usage, not Railway resources.

MiroFish templates on Railway

MiroFish Swarm Intelligence EngineMaintained by Tracklayer
MiroFish is an open-source multi-agent simulation engine. You give it documents and a scenario, it builds a knowledge graph, turns the people in it into LLM agents, runs them on si
3 deployments
Deploy
Mirofish by an independent creator
A Simple and Universal Swarm Intelligence Engine, Predicting Anything.
20 deployments
Deploy

Tracklayer templates are listed first. Other MiroFish templates follow, ordered by total deployments on Railway. How this works.

Deployment numbers as of 2026-11-20.

Questions about MiroFish

What do I need before deploying?

Two keys entered on the MiroFish service: LLM_API_KEY for an OpenAI-compatible provider, and ZEP_API_KEY for Zep Cloud, which stores the knowledge graph and agent memory. MiroFish does not support self-hosted Zep.

How do I log in?

Open the Web service URL. Use BASIC_AUTH_USERNAME (default admin) and BASIC_AUTH_PASSWORD from the Web service variables. The interface starts in Chinese; switch to English in the top bar.

What drives the cost?

Simulations make many LLM calls, billed by your LLM and Zep providers. Start with few agents and the default 10 rounds. On Railway, run a single replica and give it more memory for several simultaneous simulations.

Are the results forecasts?

No. The template notes say the results are LLM-generated simulations and should be treated as a structured brainstorm.

Are there licence notes?

MiroFish is AGPL-3.0 and the template runs it unmodified. If you change its code and let others use it over a network, the AGPL asks you to offer them your modified source.

Similar apps

Tracklayer maintains the templates marked as such; MiroFish is a trademark of its owners and this site is not affiliated with them.