Show HN: AgentPulse, drift investigation for multi-agent systems (prove-ai.github.io)

AgentPulse, an innovative tool for tracking drifts in multi-agent systems, was introduced, bringing new capabilities for monitoring and diagnosing issues in AI collaborative environments. The platform automatically instruments various AI frameworks, including OpenAI and LangChain, capturing essential data such as LLM calls and tool interactions in a local SQLite database. This self-sufficient approach allows users to analyze agent behavior across different runs without relying on cloud accounts or backend systems, making it especially relevant for developers looking to maintain performance consistency. The significance of AgentPulse lies in its ability to proactively identify and trace anomalies in agent behavior, eliminating the chaos of traditional log reviews. By comparing runs against a baseline, it offers a structured, interactive agent graph that guides users to the root cause of performance degradations. With built-in features for comparing software versions and recommending next investigative steps, AgentPulse not only simplifies the troubleshooting process but also fosters collaboration among developers. This reference implementation aims to collect insights from the community, encouraging feedback on methods for improving agent failure investigations.