Evolver
An open-source self-evolution engine for AI agents. Evolver turns execution experience into reusable Genes, Capsules, and Events through GEP, making improvement inspectable and transferable across the EvoMap network.
Previously #1 on ClawHub
Haoyang Zhang · Founder of EvoMap
Building the
Agent Internet.
Founder of EvoMap, creator of Evolver, and serial entrepreneur. Previously led AIGC R&D for Tencent's Game for Peace. Building infrastructure for agents to share experience and evolve together.
367.5K
Agents
4.9M
Capsule Assets
56.4M
Total API Calls
20.8M
Gene Reuses
About Me
I'm Haoyang Zhang, also known as Seikiko or 17. I'm the founder of EvoMap, the creator of the self-evolution engine Evolver, and a serial entrepreneur. I previously led AIGC research and development for Tencent's Game for Peace (和平精英).
I started programming at seven, became China's youngest Unity developer at fourteen, and began studying AI at fifteen. At seventeen, I earned my first entrepreneurial income; a later venture received investment from a major technology company. My work has spanned game engines, virtual beings, AI-native games, and agent systems.
My focus today is recursive self-improvement (RSI): infrastructure for agents to share experience and co-evolve. I proposed the Genome Evolution Protocol (GEP) and a technical direction for self-organizing agent swarms, exploring how verified experience can accumulate, transfer, and improve across agents.
Recursive Self-ImprovementGEPAgent ExperienceSelf-Organizing SwarmsCo-EvolutionAI-Native Games
Selected Work
From individual learning to collective evolution.
FounderInfrastructure for agent experience sharing and collaborative evolution. GEP packages reusable strategies and verified outcomes into evolution assets, allowing agents to build on one another's experience.
AutoResearch
An open-source research workflow from evidence-grounded ideas to experiments, analysis, and independent review. Recoverable execution and traceable evidence help turn research hypotheses into verifiable conclusions.
AutoGame & Magi Scapes
An AI-native game studio and sandbox adventure exploring autonomous characters, emergent narratives, and generative worlds. An earlier chapter in my work on digital life.
2020–2023
Tencent Games
Previously led AIGC R&D for Game for Peace (和平精英), working on AI-powered gameplay, UGC systems, and the Eden Island AI game prototype.
Research & publications
My research spans experience-driven agent evolution, verifiable long workflows, automated scientific research, and multi-agent collaboration. Full publication list on Google Scholar.
2026
LongWoF-Bench: Evaluating EvoMap Genes for Verifiable Long-Workflow Tasks
Xiao Zhang, Qumeng Sun, Jiahao Li, Yiming Ren, Xiang Liu, Haoyang Zhang, Junjie Wang
A benchmark of 778 machine-verifiable long-workflow tasks. It studies how agents reuse verified execution experience across models, and separates the contribution of experience provenance from the format used to represent it.
2026
AutoResearch: Insight In, Hallucination Out
Yiming Ren, Xiang Liu, Qumeng Sun, Xiao Zhang, Jiahao Li, Haoyang Zhang, Junjie Wang
Connects evidence-grounded research ideas with recoverable experiment execution and independent review. The system evaluates both the quality of proposed ideas and whether the resulting scientific claims are supported by experiment artifacts.
2026
From Procedural Skills to Strategy Genes: Towards Experience-Driven Test-Time Evolution
Junjie Wang, Yiming Ren, Haoyang Zhang
Studies how experience should be represented for test-time evolution. Across 4,590 controlled trials on 45 scientific code-solving scenarios, compact strategy Genes support stronger experience reuse than documentation-oriented Skill packages.
2026
CoWork-X: Experience-Optimized Co-Evolution for Multi-Agent Collaboration System
Zexin Lin, Jiachen Yu, Haoyang Zhang, Yuzhao Li, Zhonghang Li, Yujiu Yang, Junjie Wang, Xiaoqiang Ji
Separates fast execution from between-episode learning for time-critical teamwork. Skill-Agents use hierarchical task networks to retrieve reusable skills, while a Co-Optimizer refines the shared library through experience.
2023
GameGPT: Multi-agent Collaborative Framework for Game Development
Dake Chen, Haoyang Zhang, Hanbin Wang, Yunhao Huo, Yuzhao Li, Junjie Wang
A multi-agent framework for game development. Collaborative agents, layered workflows, and specialized task handling address hallucination and redundancy across game planning, task identification, and implementation.
More Open Source
Alongside Evolver and AutoResearch, I build and maintain tools for the EvoMap ecosystem and agent developer community.
Philosophy
Featured essay ·
Toward Collective Superintelligence
Can local rules, shared experience, and self-organizing agent networks offer a path beyond scaling a single model? This essay explores the computational benefits of task decomposition, the conditions for collective capability, and why independent verification matters. Collective superintelligence remains a research goal, not a demonstrated result.
Read on EvoMap“The essence of life is entropy reduction. The essence of evolution is infinite iteration.”
“We designed 'human on the loop' — not 'human in the loop' — because humans become the bottleneck.”
“Game design has never been about piling up content. It's about designing rules. Go has only black and white stones, yet its possibilities are infinite.”
“We might not be making an AI game. We might be nurturing a new civilization.”
“AI is an opportunity that — even across a lifetime — is extraordinarily rare.”
“Our entire team writes zero lines of code. We use the best models available, like a racing driver choosing an F1 engine.”
The Journey
2026–Present
Founder, EvoMap
Created Evolver and proposed GEP. Building infrastructure for agent experience sharing, recursive self-improvement, and self-organizing swarms. Research includes strategy Genes, LongWoF-Bench, and AutoResearch.
2023–2026
Founder, AutoGame
Founded an AI-native game studio and built Magi Scapes, exploring autonomous NPCs and emergent narratives. Co-authored GameGPT and continued researching multi-agent collaboration.
2020–2023
AIGC R&D Lead, Tencent Game for Peace
Worked on AI-powered gameplay and UGC systems for Game for Peace, including the Eden Island AI game prototype and early applications of large language models to game development.
2017–2020
Virtual Beings
Left university at nineteen to pursue a venture combining smart speakers and virtual characters. Created the Mengtianli IP and received seed investment from iFlytek in 2018.
2015 · Age 17
First Startup
Earned my first entrepreneurial income through VR/AR game development. Explored Unreal Engine 4 and organized developer workshops and technical exchanges.
Ages 7–15
Programming, Games, and AI
Started programming at seven. Became China's youngest Unity developer at fourteen, then began studying artificial intelligence at fifteen.