TokIO AI

An open-source financial research agent that treats "the data supports this" as a claim to be tested, not a vibe to be trusted.

Give it a ticker, a filing, or a plain-English trading hypothesis. It pulls real data — price history, SEC filings, a real S&P 500 snapshot — and before it will tell you a pattern is real, it runs the comparison through a permutation test, checks the sample size against a hard floor, and corrects for every other hypothesis you've asked it to test in the same conversation.

Most AI stock-chat tools will confidently describe a pattern in a handful of data points. This one is built to tell you when it can't.

View on GitHub →

What it can do

• Pull daily price history and SEC filings for any ticker, no API key needed
• Rank the real S&P 500 by trailing return, optionally filtered by sector — handles open-ended asks like "what are the best performing stocks" without you having to already know a ticker
• Test whether a technical condition actually predicts what happens next — fetch, bucket, and permutation-test in one call, not the model eyeballing raw numbers
• Correct for multiple hypotheses tested in the same session, automatically

Quickstart

pip install -e ".[dev]"
cp .env.example .env
# fill in OPENAI_API_KEY (a free key from build.nvidia.com works out of the box)
python -m tokio_ai.cli

A real example

> Test whether AAPL days that gap up more than 2% at the open tend to
  keep drifting up over the next 5 trading days, using 10 years of history.

The test found a statistically significant pattern (p=0.0050 after
correction), but in the opposite direction of the initial hypothesis.
AAPL days with gaps >2% at the open saw an average -1.31% return over
the next 5 trading days, significantly underperforming the baseline.
This suggests large upward gaps historically preceded short-term
weakness (gap-fade), not momentum continuation.