I was building AI agents and agentic systems for my freelance clients, but one of the core issues I faced was how to reliably audit AI agents, tracking execution, state drift, and complex tool call chains. As per my judgement, embedding this directly at the database layer was the right point to start, so I started building ZizkaDB in April. Under the hood, ZizkaDB isn t just a simple event logger. it s designed around specialized primitives for agentic observability: Causality Lineage DAG Tracking: Captures full execution graphs (prompts, model completions, vector retrieval contexts, tool calls, and side effects) to preserve the exact causal chain of multi-agent interactions. Point-in-Time State Engine : Allows querying the exact database snapshot and agent memory/context at any historical timestamp or step, making temporal debugging straightforward. Session Replay Primitives: Deterministically replays prior execution traces by mocking LLM/API tool outputs from historical logs, isolating code bugs from non-deterministic model outputs. I am building it as open source, and currently around 13 contributors are actively participating. One thing I realized is that agent governance and observability is a huge issue that no single mind or dev team can solve alone, which is why I went open source. Currently, it s being used by around 80 developers and the feedback is good, but I want the HN community to use it, stress-test it, and give me their honest feedback. Also, if someone would like to join in building, I would be more than happy to accept PRs!