Enki – memory for AI agents that keeps ~half as much and answers as well

- Enki is a new memory system for AI agents that compresses conversation history more aggressively than current approaches, keeping roughly half the tokens while maintaining similar answer quality. - The project benchmarks Enki against standard memory baselines (e.g., Buffer, LLMLingua) using Needle-in-a-Haystack tests and MT-Bench-style evaluations, showing comparable results with significantly less storage. - This matters because AI agents (chatbots, coding assistants, etc.) have limited context windows; better memory efficiency means lower cost, longer conversations, and more capable autonomous agents. - The repo is exploratory/experimental — not a production library — and targets researchers and developers interested in agent memory optimization.