A Comprehensive Survey of Self-Evolving AI Agents A New Paradigm Bridging Foundation Models and Lifelong Agentic Systems

Has GPT-5 already begun learning from the Twitter environment?🤔

That’s a concise yet comprehensive overview of our paper! Thanks, Rohan Paul, for the summary!

There will be an onboarding meeting for developers interested in the memory module of EAX today. Join our discord channel to find the meeting link!

3 out of 4 papers accepted at EMNLP 2025 #EMNLP2025 🎉 Congrats to Xinhao Yi, Zhaohan Meng, auss, and all the others coauthors! The remaining one (arxiv.org/abs/2505.18646) wasn’t accepted this time, but it’s still a strong piece of work, and is included in EvoAgentX!

⭐️ Star this repo and access a handpicked list of essential papers on Self-evolving AI Agents: github.com/EvoAgentX/Awes…

✨ Our survey GitHub repo hit 500+ stars in only 2 weeks🚀! Grateful for all the support 🌟!

My past grant proposals have turned into reality so quickly — just one year! Can’t imagine what the next five years will bring. 🚀 mengzaiqiao.github.io/project/ercstg/ mengzaiqiao.github.io/project/sella/ --> EvoAgentX

Thanks for sharing elvis . Check out the paper here: arxiv.org/abs/2508.07407 , and explore more in this curated list of essential papers on Self-Evolving AI Agents: github.com/EvoAgentX/Awes… 🚀 The list will be continuously updated and maintained! EvoAgentX

🚀 Excited to share that EvoAgentX, the world’s 1st open-source self-evolving framework for AI Agents, has been accepted at EMNLP 2025 (Demo Track)! It now received ⭐ 1.4k stars on GitHub: github.com/EvoAgentX/EvoA… 🎉 We’ll be presenting it in Suzhou. EvoAgentX

Hello X! 🚀 EvoAgentX v0.1.0 is live — and it redefines how agentic AI gets built, optimized, and shipped. Stop hand-wiring brittle pipelines. With EvoAgentX, agents build themselves from a single goal, then evaluate and evolve to keep getting better. * What’s inside v0.1.0 :

🚀 Revolutionizing Multimodal Knowledge Management: EvoAgentX makes AI go beyond text! github.com/EvoAgentX/EvoA… Want your agents to see images, read text, and answer questions with both? Our multimodal RAG workflow makes it real. Workflow Overview Our example demonstrates the

✨ EvoAgentX Community Call — this Sunday! ✨ We’re excited to invite you to our official open-source community meeting 🕒 When: Sunday, 9:30 AM (UK time) 📍 Where: Zoom ID: 862 8273 5663 Passcode: 396325 Over the past months, we’ve been working hard to evolve EvoAgentX from

🚀 EvoAgentX Talk – From LLM to Automated Agents: The Power of Agentic RL, join us this Sunday! 📅 Date & Time: Sep 28 2025, Sunday, 4:30 PM (Beijing Time) 🔗 Join Here: Zoom Link uofglasgow.zoom.us/j/81782288746?… or Meeting ID: 862 8273 5663, Passcode: 396325 🎙️ Session Overview

Guibin's talk will be happening in five minutes. Please come and join the discussion with Guibin.

LLM Agent = LLM + Planning + Reasoning + Tool + Memory + Action + Perception + Self-improve --- That's a much better formulation for LLM Agents.

Our survey paper’s GitHub repo just hit ⭐️1K stars! 🚀🎉 ---> github.com/EvoAgentX/Awes…

📸 New Preprint Out! If you already have a SOTA Radiology MLLM, why bother training it again? Especially when new data are scarce and GPUs are even scarcer - that's a frustrating dilemma. 🚀 We introduce a training-free and retrieval-free inference framework that boosts

It has now received 2k ⭐️ stars on GitHub!🎉 The next generation of AI Agents should be self-evolving!🍀