Show HN: Kairos – Open-source risk and fraud decisioning platform

Kairos is an open-source platform for making real-time fraud and risk decisions on events such as sign-ups, logins, orders, payments, and withdrawals. It returns an approve, step-up, review, or block outcome with a risk score and reasons, drawing on a subject’s event history and connections among devices, cards, and addresses. Configurable workflows can combine model scoring, branching, webhooks, and human review queues; case resolutions feed back as labels. The project is positioned as an alternative to commercial fraud tools, with TypeScript and Python SDKs, a CLI, an HTTP API, Docker-based self-hosting, and a hosted cloud service.

Its central idea is to let coding agents help build and improve fraud systems through an MCP server. Agents can inspect decisions, create workflows, propose model variants, and publish a winning version. Kairos compares variants using older labeled data for training, later data for validation, and the newest data for testing, reporting metrics such as ROC-AUC, PR-AUC, precision, recall, and improvement over a baseline. The platform combines Grayson, Finic’s language-based decision model - which can assess unstructured information and work with few labels - with conventional logistic regression and gradient-boosted trees trained on a customer’s own data. Grayson’s answers can also serve as features for those models. Self-hosting requires Node.js, pnpm, and a Grayson API key; the service includes its own Postgres setup.