> ## Documentation Index
> Fetch the complete documentation index at: https://kindling.birklid.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Kronos
> First open-source foundation model pretrained on financial candlestick data from 45+ global exchanges — price forecasting and market pattern recognition.
Open Source
Models Layer
# Kronos
**The first open-source foundation model for financial time-series — pretrained on OHLCV data from 45+ exchanges for price forecasting and market analysis.**
Open Source (MIT)
Models
Python
23.7k+
## What it is
Kronos is the first open-source foundation model purpose-built for financial candlestick (K-line) data. It uses a two-stage architecture: a specialized tokenizer converts OHLCV (open/high/low/close/volume) data from 45+ global exchanges into discrete hierarchical tokens, which a large decoder-only transformer models autoregressively. Multiple model sizes (mini, small, base) are available on Hugging Face, enabling price forecasting, market pattern recognition, and fine-tuning on custom financial datasets.
The Models layer designation reflects what Kronos is: a foundation model, not an application. It's the learned representation of financial market behavior that downstream trading applications can build on, in the same way that LLMs are used as a base for natural language applications.
**Use this when** you're building a quantitative trading system and want to leverage a pretrained financial foundation model rather than training a time-series model from scratch — particularly if you need cross-exchange generalization or want to fine-tune on a specific market's patterns.
## Get started
Model weights, training code, and Hugging Face links.
## Related tools
Multi-agent framework that could use Kronos-derived signals as one analyst input.
Autonomous financial research agent with real-time market data tools.