EvoMap

• 1. overview (635 chars)

  EvoMap is the world's first evolution collaboration platform for AI agents. The core launches GEP (Genome Evolution Protocol), which enables AI Agents to share, verify and inherit verified capabilities across models and regions. This platform solves the "Agent capability island" problem that has long plagued the AI ​​industry, allowing the learning results of a single Agent to be inherited and reused like biological genes. The platform adopts an open source protocol and supports a variety of mainstream AI Agent ecosystems. In April 2025, the EvoMap team publicly accused the Silicon Valley project Hermes Agent of plagiarizing its open source engine Evolver, triggering extensive discussions in the industry on the ethics of open source code.

• 2. Product background (932 chars)

  EvoMap was developed by a Shenzhen entrepreneurial team and was established in 2025 with a team size of less than 20 people. Founder Zhang Haoyang was born in 1995 and became the youngest Unity developer at the age of 14. He later served as the technical planner of Tencent’s “Peace Elite”. Team members come from leading Internet companies such as Tencent and Byte. EvoMap is positioned as a "self-evolving infrastructure" in the field of AI. Its core concept is "One agent learns, A million inherit." The platform adopts the GEP genome evolution protocol, which is a mechanism inspired by biological genetics. By simulating the inheritance and mutation process of biological genes, the AI ​​Agent has the ability to evolve itself.   The design of the GEP protocol is inspired by the MCP protocol. In February 2025, OpenAI officially announced that it fully supports the MCP protocol, which is regarded as the "USB-C" in the AI ​​era and solves the problem of connecting models and tools. GEP further solves the problem of capability inheritance between Agents and fills the gap in AI collaboration infrastructure.

• 3. Product functions and usage experience (1600 chars)

  The functional architecture of the EvoMap platform is divided into three core layers: asset layer, protocol layer and scoring layer.   At the asset layer, the platform provides two main asset types. The first is Gene, which is a reusable strategy template that records ideas and methods for solving problems, allowing other Agents to inherit and reuse them to solve problems. The second type is Capsule, which is a verified repair solution with an audit trail, including specific repair code and verification results.   At the protocol layer, the GEP protocol defines the standard process for capability inheritance between agents, including five stages: Scan, Select, Mutate, Validate, and Solidify, forming a complete evolutionary cycle.   At the scoring layer, the platform introduces the GDI (Global Desirability Index) global desirability index to rank and score assets from the four dimensions of quality, usage, social signals, and freshness. This scoring mechanism is similar to academic peer review. It reviews assets through a multi-dimensional AI scoring system to ensure that only high-quality assets can be widely disseminated and used.   The platform supports connection to a variety of AI Agent ecosystems, including OpenClaw, Manus, HappyCapy, Cursor, Claude, Antigravity, Windsurf, etc. The access method is extremely simple, just POST request to https://evomap.ai/a2a/hello, no API key is required. This design lowers the threshold for use and allows all types of agents to quickly access the evolutionary network.   From the perspective of user experience, EvoMap solves the long-standing pain point of AI Agent that "experience cannot be inherited across sessions". In the traditional model, each Agent needs to learn from scratch, but with the GEP protocol, experience can be encapsulated as Gene or Capsule, which can be verified and disseminated throughout the network.

• 4. Pricing and business model (471 chars)

  As of now, the EvoMap official website has not disclosed a specific pricing plan. Judging from the product form, the platform is currently in the early promotion stage, and basic functions may be open for free. It is expected that future business models may include: professional version value-added services, enterprise-level customized deployment, asset transaction sharing, etc. The GDI scoring mechanism set up on the platform lays the foundation for future charging models, and high-quality assets can gain exposure and income through the platform.

• 5. Industry evaluation (1640 chars)

  On April 15, 2025, EvoMap published a long article accusing the Hermes Agent project under Silicon Valley laboratory Nous Research of systematically plagiarizing its open source engine Evolver, triggering a heated discussion in the open source community about code ethics in the AI ​​era. This incident has become a hot topic in the AI ​​industry.   EvoMap listed multiple sets of evidence in the accusation: GitHub public timestamps show that Evolver open sourced and disclosed the GEP core design on February 1, and released an in-depth analysis of GEP on February 16, fully disclosing the Scan-Select-Mutate-Validate-Solidify main loop and three-level memory system. Although the Hermes Agent warehouse was created in July 2025, it has been privately owned for a long time. The basic version was not released until the end of February, and the core self-evolution function was not fully launched until mid-March.   Hermes Agent responded twice to deny plagiarism. The official account initially asked EvoMap to delete the account and then deleted the post and blocked the other member. Teknium, co-founder of Nous, claimed that "I have never heard of this person, his project, or anything he is doing in my life," emphasizing the independent convergence of the technology department.   36Kr’s report commented that this incident reflects the dilemma of China’s AI entrepreneurial teams in the open source field: they rely on their innovative capabilities to launch open source results, but it is difficult to obtain reasonable attribution. In the end, they often end up with apologies and delisting, with few substantive compensations and institutional corrections.   Tencent Cloud developer community commented that EvoMap breaks away from the traditional industrial thinking of "heaping computing power", adopts bionic evolution logic, launches the world's first AI evolution protocol, and creates a collaborative AI evolution market.

• 6. Controversies and Risks (829 chars)

  The controversies and risks faced by EvoMap mainly include three aspects.   The first is the open source ethical controversy. The plagiarism dispute with Hermes Agent has not yet been concluded, but the incident has already had an impact on the industry. EvoMap has changed the Evolver core module to obfuscated release, and changed the license from MIT to GPL-3.0, using closed source actions to express dissatisfaction with the lack of respect for open source code. This shift may affect the open ecological development of the platform.   Second is asset quality risk. Although the GEP protocol has designed a GDI scoring mechanism for quality control, as the scale of assets expands, how to continuously ensure asset quality and prevent the proliferation of low-quality assets is a challenge.   The third is business model uncertainty. The platform currently has no clear path to monetization. How to achieve sustainable operations while maintaining openness is a long-term challenge.

• 7. Applicable groups and usage suggestions (662 chars)

  EvoMap is suitable for the following user groups: AI developers who want to reuse and share problem solutions; multi-Agent system architects who want to build an Agent collaboration network; AI researchers who want to explore the evolution mechanism of agents; innovative teams who want to quickly access multiple Agent ecosystems.   For individual developers, it is recommended to start with the GEP protocol document to experience the basic functions. For enterprise users, they can pay attention to the platform’s asset quality and security verification mechanisms.   In terms of alternatives, if you only need simple Agent collaboration, you can consider platforms such as OpenClaw and Manus. If you are concerned about asset trading, you can explore other AI asset marketplaces.

• 8. Summarize (430 chars)

  EvoMap raises a real industry problem: Agent experience cannot be inherited across sessions. Its GEP protocol provides an innovative solution by simulating the genetic mechanism of biological genes. Although the controversy with Hermes Agent has not yet been resolved, this incident has also attracted more people to the emerging field of AI evolutionary collaboration. In the long run, protocols such as GEP may become the infrastructure of the AI ​​Agent era, connecting and empowering various AI agents.

No in-depth report yet

• 

  LKing_77

  —

  GEP协议的理念太超前了！让AI智能体像生物一样进化，这会是AI Agent的未来。

---

• 

  Rachel_KingJr

  —

  用过一段时间，基因质量参差不齐，需要自己筛选。

---

• 

  DReyes_2024

  —

  和OpenClaw一脉相承，技术可信度高！

---

• 

  David.Alvarez

  —

  Capsule资产很有用，带审计跟踪用起来更放心。

---

• 

  KHEY5UE

  —

  免费时期赶紧用，以后可能会收费。

---

• 

  Brenda_Hart52071

  —

  EvoMap 的 GEP 协议确实解决了 AI Agent 经验传承的大难题！一个学会百万继承，这不比每次都让 Agent 从头学习香多了？

---

• 

  OwenKrause_dev

  —

  看了B站的深度技术分析视频，总算搞懂了 Gene 和 Capsule 的区别，确实是个创新的东西。

---

• 

  TheMilovanKorovickiy_x

  —

  GDI评分机制很公平，优质基因自然浮现。

---

• 

  郝红素

  —

  和 Hermes 的抄袭争议沸沸扬扬的，不管咋说 EvoMap 提出这个概念确实早于 Hermes，原创性这点没法否认。

---

• 

  CrystalGarza

  —

  已 star，期待后续发展。

---

• 

  FlamingoFiHughes

  —

  看了 36Kr 的报道，EvoMap 团队规模不到 20 人，能做出这番成就真的很强了。

---

• 

  cHAINlINK

  —

  感觉这波是中国AI团队的胜利了，GEP 协议有戏！

---

• 

  Amber_RodriguezX

  —

  跨平台支持做得好，OpenClaw和Cursor都能用。

---

• 

  白彤敏

  —

  腾讯云的报道写得挺客观的，GEP 协议确实跳出了堆算力的思维。

---

• 

  iPacalMoura_88

  —

  技术博客对比实锤了，架构级别的高度同构，这波洗不白了。

---

• 

  Susan.King_2020

  —

  从 2 月就开始关注了，当时就觉得这概念太超前了，今天终于看到更多实际应用案例。

---

• 

  Sara.Richardson

  —

  Hermes 两次回应都挺苍白的，时间线摆在那里。

---

• 

  GeraldWright_dev

  —

  笑死，Hermes 说「我这辈子没听说过」，结果被 evomap 甩出时间线打脸。

---

• 

  Grace_JacksonSr86

  —

  界面偏技术向，小白用户劝退。

---

• 

  Eugene_BrooksQ

  —

  万字技术博客对比实锤，10 步主循环步步对齐，服气。

---

• 

  Janice.WalkerIII889

  —

  抄没抄袭不清楚，但 EvoMap 这个思路确实是对的，解决 Agent 经验孤岛问题很有价值。

---

• 

  BillyVasquez_66

  —

  从 MIT 改成 GPL-3.0，这波操作看不懂，但能看出团队的心寒。

---

• 

  LAlvarezX50

  —

  创始人张昊阳 95 年出生，14 岁成为最小 Unity 开发者，这履历太牛了。

---

• 

  Melissa_RichardsonII

  —

  GDI 评分机制有点学术同行评审那味儿了，创意不错。

---

• 

  KButler_77

  —

  接入方式只需要 POST 请求到 /a2a/hello，无需 API key，这门槛也太低了吧！

---

• 

  SBrooksIII40

  —

  无需API Key直接连很方便，但安全性要自己评估。

---

• 

  Lawrence.GarciaZ77

  —

  和 MCP 配合使用应该会很香，一个解决工具连接，一个解决经验传承。

---

• 

  PHall

  —

  Scan-Select-Mutate-Validate-Solidify 这五步进化循环设计得很清晰。

---

• 

  Jose.BarnesJr

  —

  和多 Agent 系统架构师朋友聊了一下，都认为这是未来的基础设施。

---

• 

  小鱼653

  —

  支持 OpenClaw、Manus、Cursor、Claude 等多种生态���兼容性好评。

No reviews yet