After Hermes, these self-growing Agents are the most worth pursuing!
Brothers, after the Hermes Agent became popular online, everyone is asking if there are other self-evolving AI Agents like this increasingly intelligent digital avatar besides Nous Research?
The answer is: Yes, and there has already been a wave of self-evolution in 2026.
The core of these projects captures the essence of Hermes: closed-loop learning, autonomous skill refinement, persistent memory, and self-optimization. They are no longer just temporary workers but truly long-term partners that can grow. I have searched across the internet including GitHub, arXiv, Reddit, Zhihu, and Bilibili, selecting the 5 open-source/framework projects that are closest to Hermes, ranked by similarity and practicality. Each one includes GitHub, core strengths, difficulty level, and a comparison with Hermes to help you quickly choose:
🟢
1. EvoAgentX (most recommended! Most like Hermes' “evolution ecosystem”)
• GitHub: github.com/EvoAgentX/EvoA…
• Core trick: complete self-evolution framework, supports building → evaluating → evolution closed loop. Automatically optimizes Agent workflow and skills using strategies like retrieval enhancement, mutation, and guided search. Built-in Self-Evolution Engine allows Agents to “self-upgrade” like software iterations. Also includes research papers (Self-Evolving AI Agents).
• Compared to Hermes: Hermes leans towards personal persistent memory + skill files, EvoAgentX is stronger in automatic evolution of multi-Agent workflows, suitable for complex tasks. Both can “get stronger the more you use them”, but EvoAgentX evolution is more automated (no manual patching required).
• Difficulty level: intermediate (supports Ollama, local deployment).
• Suitable crowd: developers or heavy users who want to play with “Agent ecosystem” evolution.
Hermes is a personal avatar, EvoAgentX is an evolvable legion. One of the hottest self-evolution frameworks in 2026.
🟢🟢
2. aiwaves-cn/agents (Agents 2.0) (data-driven self-evolution king)
• GitHub: github.com/aiwaves-cn/age…
• Core trick: symbolic learning + data centralization to achieve true self-evolution. Agents automatically extract experiences and update their own logic from interactions, forming a “lifelong learning” closed loop. The paper title is (Symbolic Learning Enables Self-Evolving Agents). Supports autonomous language Agents, memory and skills grow with tasks.
• Compared to Hermes: both emphasize “learning from experience”, but aiwaves is more academic + data-driven, and skill evolution is more symbolic (high explainability). Hermes' four-layer memory system is more practical, and this lifelong learning is more hardcore.
• Difficulty level: intermediate (Python framework).
• Suitable crowd: researchers/developers pursuing explainability and self-growth.
If Hermes understands you more as you use it, this one becomes smarter + traceable the more you use it.
🟢🟢🟢
3. Agent0 Series (aiming-lab/Agent0) (zero data self-evolution dark horse)
• GitHub: github.com/aiming-lab/Age…
• Core trick: self-evolution from zero data - no human-labeled data needed, evolves directly from tasks through tool-integrated reasoning + self-iteration. Supports Agent0 (language Agent) and Agent0-VL (visual language Agent), emphasizing “Tool-Integrated Reasoning” to allow Agents to grow autonomously.
• Compared to Hermes: Hermes relies on user interactions + skill files to accumulate experiences, while this one is more radical - it can self-evolve starting from “zero data”. Both persist in learning, but Agent0 is more suitable for exploratory tasks.
• Difficulty level: intermediate to advanced.
• Suitable crowd: geeks who want to experiment with “pure autonomous evolution”.
Hermes is a companion-style evolution, this is a self-reliant type from 0 to hero.
🟢🟢🟢🟢
4. Tencent/SelfEvolvingAgent (WebEvolver, etc.) (produced by Tencent AI Lab)
• GitHub: github.com/Tencent/SelfEv…
• Core trick: WebEvolver! Web Agent self-improvement + co-evolution world model. Agents continuously self-optimize in real environments, build dynamic world models, and achieve self-evolution for long-term tasks.
• Compared to Hermes: Hermes is general + multi-platform, Tencent specializes in web/environment interaction self-evolution, with a world model that makes it “stronger as it fights” in dynamic scenarios. Enterprise-level endorsement, high code quality.
• Difficulty level: intermediate.
• Suitable crowd: those working on browser automation and web Agents.
Hermes is an all-round evolution, this is a specialized evolution machine for web battlefields.
🟢🟢🟢🟢🟢
5. CharlesQ9/Self-Evolving-Agents + Awesome list (research + tool collection)
• GitHub: github.com/CharlesQ9/Self… (main project) + github.com/EvoAgentX/Awes… (most comprehensive)
• Core trick: not a single framework, but a panoramic research on self-evolving Agents, including memory evolution, reflex capabilities, lifelong learning, etc. The Awesome list includes all top papers, benchmarks, and open-source projects from 2025-2026, providing a one-stop understanding of the ecosystem.
• Compared to Hermes: Hermes is practical, this series is “theory + toolbox”. Must see for those wanting to delve into self-evolution mechanisms!
• Difficulty level: low (research-oriented) → medium (play along with the code).
Hermes is a single soldier, this is a self-evolving all-in-one solution + roadmap.