cognicore-dev/cognicore-env ? reverse-engineered prompt

Build me a Python tool for AI agents that gives them persistent memory across runs, so they can remember past tasks, search old experiences by meaning, and use that context the next time they work. I want it to feel like a small memory operating system for agents, not just a chat log.

It should also learn from mistakes, so after a failed run it can reflect and improve the next attempt, and it should block obvious prompt injection or jailbreak attempts. If possible, include a way to replay past agent decisions and branch from an earlier step, plus a simple demo agent and a few example environments I can run locally.

I’d like it to work without requiring API keys for the basic features, with optional support for common LLMs and a simple local dashboard or server for trying things out. If you need to look up any current docs online, go ahead.