MiroShark
VerifiedFramework for LLM agents simulating social interactions on Twitter, Reddit, and markets.
What is MiroShark?
MiroShark is an open-source social simulation framework built around LLM agents that operate inside replicas of Twitter, Reddit, and a prediction market.
Agents exchange messages and make decisions at each hourly tick, with full support for loading custom scenarios, creating counterfactual branches, and invoking tools per agent via the MCP interface.
It is intended for researchers and developers who need repeatable, controllable environments to study collective agent behavior without relying on live external platforms.
Capabilities
What you can build with MiroShark
Scenario testing
Load predefined or custom scenarios to observe how groups of agents respond to specific events or policy changes across the simulated networks.
Counterfactual analysis
Branch a simulation at any point to compare outcomes when different agent decisions or external conditions are introduced.
Tool-augmented agent studies
Examine how individual agents leverage MCP tool calling to gather information or execute actions inside the social and market environments.
Install MiroShark
git clone https://github.com/aaronjmars/MiroShark.git && cd MiroShark
cp .env.example .env
# Paste your OpenRouter key into the LLM_API_KEY / SMART_API_KEY /
# NER_API_KEY / OPENAI_API_KEY / EMBEDDING_API_KEY slots (same key,
# 5 places). Default lineup is Mimo V2 Flash + Gemini 3 Flash.
./miroshark- 1Clone the MiroShark repository from its source.
- 2Install required Python dependencies listed in the project files.
- 3Create or select a scenario configuration file.
- 4Launch the simulation with the chosen scenario and tick settings.
- 5Review generated logs and branch outputs for analysis.
Works with
MiroShark: pros & cons
Pros
- +Unified simulation of multiple social platforms plus a market
- +Built-in support for scenario branching and counterfactual runs
- +Per-agent tool calling through MCP
- +Hourly discrete time steps for controlled pacing
Cons
- –Platform coverage limited to the three provided environments
- –Requires compatible LLMs and MCP setup for full functionality
- –No real-time external data integration
Frequently asked questions
Agents operate inside simulated versions of Twitter, Reddit, and a prediction market.
User reviews
Verified reviews from the community shape this listing's rating.
Loading reviews…
Similar agents
Other agent frameworks options worth comparing.
