What Kairos is
Kairos is a library that consumes an options-market data feed and emits the analytics an institutional trading or risk system needs. It runs in your process, owns no persistence, and exposes a single fluent client.
The shipped surface covers fourteen analytics — options Greeks, CBOE-method volatility surfaces, OHLCVC bars, Lee–Ready aggressor-classified trades, options-flow stratification, stock-flow alerts, order-flow toxicity (VPIN), gamma exposure (GEX), 52-week price bands, unusual-volume detection, implied-volatility change alerts, historical realised volatility, block-trade detection, and a per-contract moneyness slice.
Every analytic obeys three contracts:
- Replay parity — the output stream produced by replaying a recorded tick capture equals — frame for frame, field for field, watermark for watermark — the stream the same analytic would have emitted live.
- Lock-free dispatch — a single OS thread drains the ingest path; per-symbol state cells are owned by that thread; the dispatch path takes no locks in steady state.
- Trust-boundary validation — every wire decoder, every prewarm decoder, and the quote-grid admission path fails closed on
NaN, ±Inf, negative size, or out-of-range date.
How Kairos is delivered
Three execution modes, one client, one output type per analytic:
- Live — subscribe to typed output streams driven by the upstream websocket session. See Live streaming.
- Historical — query a date range across a contract universe. Each
(contract, date) unit emits exactly one terminal watermark and a per-unit verdict. See Historical query. - Replay — drive any analytic from a recorded
TickDataset (Vec, iterator, or the tick parquet schema). See Dataset replay.
Status
The current release is v3.0.0. See the changelog for the release history and the v2 → v3 migration guide for the breaking changes.