Open source

Simulate how the world reacts.

Hundreds of AI agents with real personalities, grounded in live market data. They post, debate, predict, and self-calibrate against reality.

View on GitHubDocumentation
mirowhale
$ mirowhale ingest --all
[crypto]BTC/USDT$70,522+2.1%
[crypto]ETH/USDT$2,143-0.4%
[stock]NVDA$172+3.2%
$ mirowhale simulate --agent 100

Generating agent... 100 created in 4.2s

Running simulation... tick 1/144

Marcus Chen: "BTC just broke $71,200. Watching $72K."

Elena Kozlov: "Risk-adjusted ETH return concerning."

Simulation complete. 847 post, 23 prediction.

Feature

A living world model.

Real-world data

Live stream from Twitter, Polymarket, crypto exchange, stock market, and news wire. Agent react to what is actually happening, not hypothetical scenario.

Deep personality

Each agent has a unique backstory, bias set, and posting style. Mirror agent modeled on real account. Archetype calibrated to real sentiment distribution.

Self-calibrating

Agent make prediction. Real outcome stream in. Brier scoring separate signal from noise. Accurate agent gain influence. Bad predictor fade out.

How it works

Five stage. One pipeline.

01

Ingest

Twitter, Polymarket, crypto, stock, news

02

Extract

LLM-powered NER, knowledge graph, embedding

03

Generate

Deep persona calibrated to real sentiment

04

Simulate

Agent post, debate, shift opinion each tick

05

Score

Brier scoring, influence weight evolve

Data

Grounded in reality.

Twitter / X

Apify

Tweet, engagement metric, real-time sentiment

Reddit

Apify

Hot post from crypto and finance subreddit

Polymarket

pmxt

Prediction market odd from 7 exchange

Crypto

ccxt

BTC, ETH, SOL from 100+ exchange

Stock

Polygon

AAPL, NVDA, TSLA, MSFT, META

News

GDELT

Global news, 100+ language, real-time

Agent

Real personality.

MC

Marcus Chen

Crypto Trader

27, Bay Area. First BTC trade at 19. Burned by the 2022 crash but doubled down. Post technical analysis thread at 2am.

Accuracy
72%
DAO

Dr. Amara Okafor

AI Researcher

34, Lagos to London. Published on emergent behavior in LLM system. Skeptical of crypto hype. Write long, measured take.

Accuracy
68%
EK

Elena Kozlov

Risk Analyst

41, former Sberbank quant. See everything through a risk-adjusted lens. Will never post without citing data. Bearish by default.

Accuracy
81%

Open source

MIT licensed. Fork it.

The entire codebase. Every prompt. Every algorithm. 151+ file, zero secret.

FastAPINext.jsSupabasepgvectorGroqClaudeOpenRouterccxtpmxtApifyGDELT
Star on GitHub