Ayan from Finic launches Kairos, an open-source alternative to Sift, Sardine, and Riskified. Kairos addresses two major challenges in traditional fraud decisioning: the lack of labeled data and the need for constant model updates as fraudsters evolve. Traditional models like XGBoost rely on quality labeled data, which many businesses lack. Kairos introduces "System One" decision models that allow plain language questions, such as "Is this a throwaway email," bypassing the need for large datasets. It also captures unstructured fraud aspects that gradient-boosted trees struggle with.
Kairos tackles evolving fraud tactics by introducing a Model-Context-Protocol (MCP) server. This allows AI agents like Claude Code, Cursor, or Codex to define decision workflows, train models on user data, and select the best-performing variant. This reduces the need for a large data science team to maintain detection logic.
The platform offers real-time decisions, workflows, queues, and experiments. Developers can self-host with Docker or use the hosted version. SDKs are available for TypeScript and Python, along with a CLI and HTTP API. Kairos aims to act as an AI fraud ML engineer, continuously tuning models on product data.
*Key entities bolded for emphasis*
Source: Hacker News · Summarized by HeadlinesBriefing