Kyros is OS-level memory infrastructure for AI agents. Instead of re-sending full chat history on every call, Kyros pages, scores, and evicts memory like an operating system manages RAM using real decay curves, not just a vector DB with a TTL. Open-source core, 5 framework integrations (LangChain/CrewAI/AutoGen/LlamaIndex/Semantic Kernel), Merkle-signed writes for auditability. Built by backend engineers tired of watching context bloat eat compute budgets.
No reviews yetBe the first to leave a review for Kyros
Maker
📌
Hey PH! Jai here, solo founder of Kyros. 👋
I’m a backend engineer who got tired of watching expensive LLM calls burn cash re-processing context the agent had already seen five messages ago. No eviction, no paging just the entire history resent forever. It felt like we were ignoring decades of OS research.
So I built Kyros. It’s not just another vector DB; it applies actual memory-management principles (paging, working sets, decay) to the AI agent stack.
What makes it different:
● True Memory Types: Episodic, semantic, and procedural memory (not just one big vector blob).
● Natural Decay: Memories fade using the Ebbinghaus curve so your context window stays lean.
● Tamper-Evident: Every write is Merkle-hashed by default.
● Drop-in Ready: Works with LangChain, CrewAI, AutoGen, LlamaIndex, and Semantic Kernel.