Hey everyone I'm Kaushal, an engineer working on autonomous agents, agent memory, and multi-agent systems. Over the past few months I've been building an open-source project called CogniCore focused on a question that kept bothering me: Why do agents keep repeating the same operational mistakes? Instead of only storing conversation history, CogniCore stores: Failures Reflections Successful recovery paths Replayable execution traces A few findings from my experiments: • AutoLearner improved from 38% → 95% when failure memory was enabled. • In ablation studies, adding a reviewer agent actually reduced performance while increasing token usage (+9,642 tokens). • In an OpenYabby integration experiment, baseline workflows repeated failures, while experience-memory workflows successfully recovered on follow-up tasks and enabled cross-agent knowledge transfer. Current areas I'm exploring: Experience memory for agents Replay and debugging systems Cross-agent learning Multi-agent orchestration Production AI infrastructure Would love to connect with others working on agent reliability, memory systems, LangGraph/LangChain workflows, and long-running autonomous agents. Pip install cognicore-env (we have crossed 6k downloads ) https://github.com/cognicore-dev/cognicore-my-openenv