Kōan

Kōan is an observability platform that lets users watch the internal processes of AI agents as they run. It captures and displays the agents’ reasoning steps, the calls they make to external tools, and the decisions they reach, providing a real‑time view of their operation. The system supports multiple large‑language‑model providers, including Anthropic, OpenAI, Gemini, and any custom LLM that can be integrated, allowing developers to monitor agents across different back‑ends from a single interface.

The platform is aimed at developers, researchers, and engineers who build or maintain autonomous AI agents and need insight into how those agents arrive at their outputs. By exposing the sequence of thoughts, tool invocations, and final choices, Kōan helps diagnose unexpected behavior, verify correctness, and improve transparency in complex, multi‑step workflows.

As an experimental project, Kōan focuses on delivering detailed, time‑synchronized visualizations of agent activity without imposing a specific workflow, making it a flexible addition to existing AI development pipelines.