# ZizkaDB

A self-hosted operational database for auditing AI agent behavior, recording each step's causal chain back to the user message so a team can answer why an agent called a given tool.

ZizkaDB is a self-hosted audit trail database for AI agents, aimed at making agent behavior traceable and compliant with regulation such as the EU AI Act. It logs each agent step with a parent ID linking it to the step that caused it, so a team can walk back from a surprising tool call or answer to the user message that triggered it.

A dashboard and a command-line lookup surface the causal chain for a given event, and the project ships SDKs for Python and TypeScript plus integrations for LangChain, CrewAI and LiveKit voice agents, alongside a REST API and an MCP server. This repository is the open-source self-host stack; an operator console and VPC deployment are offered separately as a managed cloud product.

Website: https://db.zizka.ai