Launching today
Eros

Eros

Episodic memory for the autonomous AI agents

2 followers

Visit website
AI Agents are stuck in a "Permanent Present." Every time you ask an agent to perform a recurring task— it starts the execution from scratch. This makes agents slow, expensive, and unpredictable. The Solution: The Episodic Plan Cache is the memory layer that gives AI agents a "Reflexive Memory." Instead of generating a new execution graph for every request, it indexes successful past outcomes. When a similar intent is detected, it hydrates a proven plan in milliseconds instead of re-generates.
Eros gallery image
Free
Launch tags:Open Source•Artificial Intelligence•GitHub
Launch Team
Unblocked AI Code Review
Unblocked AI Code Review
High-signal comments based on your team's context
Promoted

What do you think? …

mahika jadhav
Maker
📌
I was building a behavioral nervous system, basically was trying to bring J.A.R.V.I.S to life. For that project I developed a lot of sub-systems, with the goal of re-building the human brain exactly as it is. The thought was if I replicated all of the systems, the AI might feel real? During that I created learning loops and memory systems for intentional growth through each new experience. Later on when I started browsing the open source projects, a huge thing I came across was the lack of active learning or using of the experience of AI agents. Founders described stateless agents as reliable and more predictable but my theory was opposite. Stateless agents are even more unpredictable as they completely rely on the LLM output, it has no thinking of its own. LLM outputs vary each time even with the same prompt. That is the long winded story of why this was developed and why I think it is needed.