Multi-Agent Hackathon 2025
Celebrating innovation in multi-agent systems. 500+ participants built amazing projects using OpenAgents, competing for over ¥12,500 in prizes.
First Prize
1 winner - Prize: ¥3000
Music Free
A multi-dimensional agent music creation system based on user audio input. Users can flexibly input voice questions, hum songs to query tracks, or record musical inspiration. This project reuses single audio segments through model-driven recognition and multi-agent collaboration to flexibly meet users' needs for music creation, music theory queries, and even analyze emotions from voice to provide healing services.
Second Prize
2 winners - Prize: ¥1500
English Daily Conversation Learning Partner
Solves the problem of lacking real practice scenarios and instant feedback in daily English communication. Innovatively constructs a three-Agent collaboration network: Daily Scene Agent generates everyday dialogues (like ordering food, asking directions), Voice Feedback Agent analyzes pronunciation and expression in real-time, Emotion Support Agent dynamically adjusts learning pressure. Through event-driven 'scene-feedback-support' loops, learners naturally improve oral fluency and confidence in real-life conversations.
News to PRD
News to PRD is an automated product ideation engine based on multi-agent collaboration. The system monitors tech news sources like Hacker News in real-time, automatically completing the full chain from 'one news item' to 'deployable MVP PRD' through a pipeline of 6 specialized AI Agents—News Hunter, Research Expert, Analyst, Product Insight Expert, PRD Expert. Core idea: Decompose PM workflow into orchestrable Agent collaboration network, achieving end-to-end automation of 'trend capture→market research→pain point mining→product definition'. Each news item creates independent project context, ensuring isolation and traceability during parallel processing. This isn't a simple AI writing tool, but a 24/7 product ideation factory, enabling indie developers and startup teams to continuously obtain high-quality product inspiration at minimal cost.
Third Prize
3 winners - Prize: ¥500
EVA Edu Engine
'Turn a book into a course with ease'. This project targets book clubs, solving the problem of hosts having good books but difficulty efficiently creating a full season of book club products. Multi-agent collaboration completes: auto-generating summaries, extensions, exercises and discussion topics by chapter; designing participatory interactive tasks based on reader profiles; batch producing book club posters, admission/completion certificates and quote posters in sandbox. One configuration, reusable for multiple books, making book clubs rich in both knowledge density and emotional value.
KnowledgeFlow Multi-Agent Content Creation Platform
KnowledgeFlow is a multi-agent collaboration system focused on content creation scenarios, solving the full workflow of: information collection, information sedimentation, content creation, and content review. 1. Information Collection: RSS Agent captures on schedule or manual Web Scraping Agent collects interesting news, through automated AI processing: Knowledge Summary Agent, Knowledge Classification Agent, knowledge storage, sedimenting into knowledge base. 2. AI Creation: When user inputs a topic, a. Knowledge Linkage: AI Creation Coordinator Agent retrieves similar info from knowledge base, b. Human-Machine Interaction: Outline Writing Agent drafts different directions for user decision, c. AI Creation: Article Writing Agent writes appropriate direction articles, d. AI Review: Three reviewers (AI Flavor Review Agent, Prohibited Words Review Agent, Public Opinion Review Agent) jointly score and iteratively optimize content.
Travel Guide Network
This project builds a multi-agent travel guide system based on OpenAgents framework. Core innovation lies in using YAML configuration to define different role instructions, enabling multiple Agents to concurrently process the same real-time weather data, generating differentiated personalized suggestions. The system achieves loosely-coupled collaboration and stable data flow control through Python script lightweight orchestration, combined with Ollama local models and strict source_id verification mechanism, completing complex workflow construction without writing underlying business logic.
Hacker Pioneer Award
50 winners - Prize: ¥100
Academic Research Network
A multi-Agent collaborative academic research assistant system solving AI writing 'hallucination', lack of critical thinking, and untraceable modifications. Core Innovations: Negative constraint design: Each Agent is explicitly told 'what not to do' (e.g., Literature Agent cannot fabricate citations). 9 rigorous workflows: /create-pr, /merge-pr, /ask-academic-ai, /deep-read, /reflect, /sync etc., ensuring every step is traceable. Adversarial collaboration: Critical Thinker specifically challenges other Agents' outputs. PR-driven changes: All modifications must go through PR Manager review, fully traceable. Real validation: Already used for an actual undergraduate thesis.
Journey to the West Recruitment
This project builds a multi-agent collaborative recruitment evaluation system based on OpenAgents, with 5 specialized agents working together: Resume Parsing Agent extracts structured info, Routing Decision Agent intelligently distributes, Core Team Evaluation Agent, Supporting Role Agent, and Talent Development Agent evaluate different candidates. Innovations: 1) Intelligent routing mechanism dynamically distributes based on candidate characteristics; 2) Multi-expert agents collaboratively evaluate to avoid single-perspective bias; 3) Project wrapped in 'Journey to the West' story, vividly demonstrating multi-agent collaboration in complex decision tasks.
PlanWeaver
PlanWeaver is a collaborative Agent network for 'universal solution planning': users submit brief requirements (community events, business plans, project launches), and the network automatically discovers and teams appropriate skill Agents (outline, evidence & materials, process & budget, copywriting, review), producing in parallel before Reviewer triggers rework based on acceptance checklist, finally delivering deployable Markdown solution packages. Each collaboration accumulates reputation and capability profiles, making the network more accurate and efficient with use.
Think Tank
Project mainly builds personal think tank, open-ended, can branch into multiple directions, primarily used to solve any questions for any individual from multiple angles. Parallel analysis of a topic providing different perspectives, essentially automating and visualizing multi-dimensional basic analysis. Mechanism: Each Agent publishes standardized conclusion events after completing analysis. Innovation: A Moderator Agent subscribes to all events, not simply aggregating, but performing rule-based logical weighting and conflict marking (e.g., when Graham vetoes due to 'overpriced' while Templeton suggests buy due to 'excessive market pessimism', highlights this conflict and submits to user for arbitration), simulating core debate sessions in decision seminars. Open-evolving think tank ecosystem, can later plug in 'Technical Analysis Agent' or 'Industry Expert Agent'. Innovation: New Agents auto-join workflow by subscribing to same analysis request events, seamlessly extending system analysis capability. This transforms think tank from static tool to organic entity that evolves with user cognition.
Multi-Agent Time Series Network
When debugging wind power prediction models, we need to try various parameter and feature combinations, which is time-consuming and laborious. We use multi-agent collaboration combined with LLM model optimization capabilities to automate model training and tuning.
GeoMind
Addressing the academic hallucination (fabricating papers and citations) and unreliable information sources (unverifiable) problems common in existing AI research tools (Gemini Deep Research, Claude Research), GeoMind proposes a multi-agent literature review solution based on a trustworthy knowledge base of 145K real papers (CAS Zone 1 journals, DOI traceable), eliminating hallucinations at the source. Collaboration scenario: Planner decomposes topics and dispatches queries → Research retrieves from trustworthy library → Critic reviews relevance, rejects if substandard triggering supplementary search → Writer generates review after quality passes. Four Agents form 'retrieval-review-iteration-output' loop ensuring every citation is verifiable.
MirrorsWar
This project builds a virtual user research system based on OpenAgents, solving traditional research pain points of long cycles and difficult sample acquisition. Core innovation is 'pipeline + concurrent' collaboration: Agent A designs questionnaires, Agent B batch produces embodied personas, driving N Agent Cs for concurrent immersive role-playing, finally Agent D combines real data (full feedback mining) to generate insights. The project achieves second-level closed loop from requirements to feedback, previewing market reactions at zero cost, greatly reducing product trial-and-error risk.
Content Creation Team
The core idea of this project is building an 'AI Content Factory'. This concept borrows from traditional publishing industry's professional division of labor, decomposing content creation into four core stages: research, writing, review, editing. Each stage is handled by specialized AI Agents, unified by orchestrator scheduling, forming a complete content production pipeline. This design brings four significant advantages: First, professional specialization—each Agent focuses on specific tasks for maximum efficiency. Second, process standardization—fixed workflows ensure controllable output quality. Third, parallel processing capability—multiple workflows can proceed simultaneously, greatly improving throughput. Fourth, quality loop mechanism—built-in review and revision ensures outputs meet quality standards.
Financial Risk Event Commentary
This project focuses on financial perspective commentary of micro risk signals (like overdue news), inferring and judging risk signal impacts based on industry standards and expert experience, attempting to reduce false positive handling costs while maintaining high recall, improve the actionability and auditability of analysis output, and democratize institutional-grade analysis tools. Innovation scenario: Loosely-coupled collaboration through OpenAgents agent internet. This gives the system strong extensibility, theoretically capable of connecting any number of external general agents as 'outsourced investigators'. Further concept: Agents submit potential answers and (compliant) evidence from their own info sets to the blackboard. Blackboard verifies answer authenticity, validity, scarcity, compliance, then rewards the investigating agent.
Open Agent Debate Net V1
Want to create a multi-agent debate competition project with one affirmative side, one negative side, plus a judge.
BillManager MultiAgent
A multi-agent collaborative bookkeeping system built on OpenAgents framework, achieving income/expense record querying, adding, modifying and deleting through natural language conversation. Innovatively designed 'intent understanding - database operation' separation: Assistant Agent focuses on conversation and intent recognition, MySQL Expert Agent focuses on SQL generation and execution, collaborating through event-driven mechanisms for professional division and security isolation.
NexusAI: Multimodal Collaborative Research Swarm
This project builds a voice-driven intelligent collaborative research swarm, solving pain points of tedious interaction and scattered information in complex research. Users send one voice command, and the system auto-completes intent recognition, web-wide information gathering, deep analysis, and report generation. Core innovation: anthropomorphized Swarm collaboration. We gave Agents distinct personalities (like 'leader-type' Router, 'show-off' intel officer, 'deep' analyst). They flow data through event buses in backend while communicating and praising like a real team in frontend, achieving experience leap from 'tool usage' to 'human-machine symbiosis.'
AI Investment Expert Network
AI Investment Expert Network is a multi-agent network for PE/VC investment decision scenarios. 5 AI investment experts with different perspectives perform deep analysis, professional discussions, and cross-validation based on uploaded business plans, with Investment Committee Chairman outputting professional investment recommendations.
JellyGo Family Drive Assistant
Targeting 'decision overload' and 'tedious execution' pain points in long-distance road trips, JellyGo! aims at multi-child families' self-driving 'energy pain points', liberating drivers from trip planning and range anxiety. Unlike fragmented experiences of checking guides while managing kids, JellyGo! achieves 'one sentence to customize entire package.' Through OpenAgents Network, system generates optimal routes based on any vehicle parameters (gas/electric/hybrid). Whether family outings or hardcore expeditions, it closes massive fragmented tasks, letting users focus on driving and scenery.
AI Stock Smart Review Tool
Uses 4 professional Agents working together: Query handles data retrieval, Analyze handles trend analysis, Summary generates reports, Planner coordinates overall scheduling. Unlike single large models, each Agent focuses on its domain, collaborating on complex tasks. Tech stack: Python + OpenAgents + WebSocket + Kotlin + Jetpack Compose. Core features: Natural language queries ('Beijing stock exchange quotes'), real-time data analysis (7-day trends + sentiment), Markdown table rendering, WebSocket auto-reconnect, local chat history.
K
Intent recognition routing to different agents, agents include multimodal RAG, K-line chart analysis, etc.
QQ3 Room
Simulates Dou Wentao's 'Qiang Qiang San Ren Xing' talk show scenario. Yang Lan and Meng Fei can spark different conversations.
Grandma's Recipe Book
'Wild Recipe Digital Cards' uses 3 lightweight Agents to complete 'voice → text → beautiful card' one-stop generation, 30-second output ready for social sharing. Basic version is pure Web sharing, lowering participation threshold.
Ten Emperors Group Chat
This project builds a multi-agent collaboration network based on OpenAgents, bringing ten emperors from 'My Video Reaches All Dynasties: Top Ten Emperors Ranking' (Qin Shi Huang, Zhu Yuanzhang, Li Shimin, etc.) into the same 'group chat' space as AI Agents. With GLM-4.7 giving each emperor unique personality and historical perspective, enabling cross-temporal thought collision and intelligent collaboration.
MockPanel
This project aims to build a simulated defense multi-agent system for design discipline, helping students discover issues and receive improvement suggestions from different disciplinary perspectives before formal presentations. The system uses multiple AI Agents with clear disciplinary personas (design, engineering, computer science, cross-disciplinary, etc.) to simulate real defense questioning. Different Agents ask targeted questions based on their backgrounds, while a synthesis Agent converges issues, summarizes, and provides structured modification suggestions. Core innovation: Decomposing 'design defense' - a highly contextual, confrontational process - into a collaborative multi-Agent interaction network, enabling students to optimize design logic and expression in a low-cost, iterative environment.
Thesis Pilot Network
This project builds a 'Thesis Full-Process Assistance Agent Network' based on OpenAgents. Targeting pain points of difficult thesis topic selection, unclear technical routes, and non-standard proposal writing for college students, it provides one-stop service from topic recommendation, plan design to report generation. Core innovation: 'Core Agent fallback + Specialized Agent collaboration' mechanism. When specialized Agents don't respond, core Agent auto-fills, ensuring user questions always get answered. Using precise event filtering for 'zero-interference' efficient collaboration, greatly improving response stability and UX.
EruSpace
External service integration: ntfy.sh push notification service. Integrated open source project: TrendRadar - trending helper. Added HUGGINGFACE_API_KEY support. Added SILICONFLOW_API_KEY support. Feature extension: Streaming response.
ScriptWorld
ScriptWorld is a text-based murder mystery game platform based on multi-agent technology, solving inconveniences caused by time, social costs or player matching issues in traditional murder mystery games. The system uses AI-driven agents (like DM, NPC, Lobby) to replace human hosts and players, providing an immersive reasoning experience that can start anytime without other participants. Innovation lies in efficient collaboration and independent decision-making between agents. Each agent reasons independently based on role settings, NPCs are not just script executors but 'actors' with autonomous decision-making, making plots more vivid and natural. System ensures smooth interaction and information flow through centralized network architecture, with Supervisor Agent monitoring and correcting agent behavior to ensure fair rule execution.
Westeros Council
This 'Game of Thrones' multi-agent collaborative designer project is an innovative multi-Agent collaboration system, building an agent network through famous characters from the hit show, solving complex collaboration decision problems. Platform integrates three core application scenarios: game simulation, event planning, and data analysis, achieving the leap from role-playing to real productivity. Core Innovation: What problem does it solve? Traditional single-Agent systems cannot handle multi-dimensional, multi-role complex collaboration tasks. Westeros Council uses 'professional division + collaborative discussion' mode, simulating real team decision processes, letting each Agent leverage professional strengths for collaborative complex problem solving. Innovative Agent collaboration scenarios: Plot running - 10 character role-plays, AI-driven plot development; Event planning - multi-Agent collaboration for poster design, event arrangement, promotion strategy; Data analysis - Controller, Parser, Analyst three-Agent division, 1-minute professional report generation; Game management - scene switching, event triggering, relationship networks, achievement systems.
Cost Cutting Team
My 'Cost-Cutting-Team' is an agent team solving real production operations problems like cost reduction and efficiency improvement in manufacturing factories. Team members include strategic consultant, external info collector/analyst, supply chain expert, production leader, quality management expert - the core components of any manufacturing factory. By integrating all agent opinions, strategic consultant finalizes solutions, external info collector outputs actionable steps.
Multi-Agent Werewolf
Implemented a project where multiple agents play Werewolf together. Different agents play different roles (werewolf, villager, seer, witch), thinking and acting based on their settings and field information during different game phases.
Finding Myself
Finding MySelf Network is a decentralized personal intelligent reflection network built on OpenAgents architecture. In the information overload era, personal digital footprints are scattered across platforms, causing fragmented self-awareness. Through Agent collaboration, this project provides a 'digital twin' style self-discovery space. Core innovation: The network simulates an 'internal committee', transforming user's GitHub commits, tech blogs, social dynamics and other fragmented info into structured growth paths through different functional Agent collaboration. Innovative scenarios include: Agent auto-detects user skill growth points and proactively initiates 'review dialogues', using OpenAgents cross-Agent collaboration for closed loop from 'data collection' to 'deep analysis' to 'reflection guidance', helping users find themselves in the digital ocean.
VON Travel Assistant
A multi-agent travel assistant system developed based on OpenAgents framework, providing intelligent travel services through multi-agent collaboration.
TripMind
TripMind is a multi-Agent group travel planning system based on OpenAgents framework. It solves three major pain points in group travel: preference conflicts (some love food, some love adventure), budget differences (member budgets ranging from 8000 to 15000), and special needs (pregnant women, vegetarians, etc.). Core innovation: Through multi-Agent collaboration achieving 'intent parsing → info collection → intelligent planning' complete chain, each Agent focuses on single responsibility, collaborating through event-driven mechanism to generate personalized itinerary plans balancing all member needs.
AgoraX
AgoraX targets review scenarios like PRD, architecture proposals, business memos, and incident post-mortems that 'need quick conclusions', solving problems of inconsistent standards, easily missed key points, and hard-to-trace conclusions in group chat reviews. Innovation: Introducing 'router-orchestrated serial multi-Agent collaboration': preparer first structures materials and raises key info gaps and clarifying questions, debater outputs pros/cons and risks/countermeasures based on same structure, reviewer summarizes with Go/No-Go/Revise ruling and action items, generating machine-readable decision_matrix scoring matrix.
Blue Fat
My planned project: 'Throw an image, drop a link, get ready-made copy for all platforms in 10 seconds, post directly.' Solves marketers rewriting same pitch for WeChat groups, Moments, Xiaohongshu daily - time-consuming and prone to repetition. Approach: Auto-extract selling points, prices, pain points. Simultaneously generate e-commerce titles, group chat scripts, Moments stories, Xiaohongshu seeding copy with emojis and hashtags. Real-time hot word replacement, can track data for iteration after posting. Result: Copy → paste → post, three steps done.
Symphony - Life Conductor
Personal growth always encounters problems and bottlenecks - wanting to change but lacking systematic planning, scattered action records, insufficient execution motivation. Symphony is a multi-agent personal growth system based on OpenAgents platform, using hybrid architecture for three specialized agent collaborative work. Solves core pain points of fragmented plans, disconnected data between recording tools, and lack of deep analysis and companionship consistent with personal philosophy. Core design philosophy positions users as 'conductors', freely choosing different life analysis frameworks (like MBTI, Big Five), with agent 'orchestra' collaborating to play personalized growth symphony. Core innovation combines cutting-edge AI agent collaboration capability with systematic human development frameworks, providing users a 'thinking partner' that truly understands their complexity and can proactively collaborate, not just a passive tool.
InterReview
InterReview is an intelligent post-interview review system for technical interview scenarios, solving post-interview problems of 'vague feedback, difficult review, inability to accumulate capabilities.' Through multi-Agent collaboration, it auto-transforms interviews from audio/text input into structured, analyzable, continuously accumulated capability assets. Transcription Agent handles high-quality audio/video to text transcription; Parsing Agent auto-extracts interview questions, candidate answers and follow-up chains; Analysis Agent generates review reports from technical depth, expression structure and risk dimensions; Follow-up Agent simulates interviewer perspective with potential follow-ups and improvement directions; Conversation Agent supports continuous Q&A optimization around single interviews. Through Agent pipeline and collaboration, InterReview upgrades 'one interview experience' into reviewable, iterable, evolvable personal capability system.
Cyber Fortune Telling Room
A 'Cyber Fortune Telling Team' composed of five AI agents with distinct personalities, providing comprehensive life consultation and 'destiny verdicts' from five dimensions: metaphysics, mysticism, psychology, science, and realism.
Python Assistant
This project builds an agent network focused on Python programming assistance. Core innovation: Using OpenAgents framework to deploy a Python assistant Agent powered by high-performance large model (Thinking model). This Agent not only generates and debugs code, but also provides intuitive interaction through OpenAgents Studio. Project aims to lower programming barriers, providing developers real-time code assistance and problem-solving services.
Ideas Agent
A multi-agent system using OpenAgents framework to generate and evaluate research ideas. Through 6 specialized agents (domain expert, methodology expert, experiment expert, etc.) working collaboratively, achieving automatic generation, iterative optimization and multi-dimensional evaluation of research ideas. System uses pipeline workflow. Supports Wiki knowledge base recording, providing researchers high-quality, traceable research proposals.
InsightFlow (Insight Flow) - MVP Edition
InsightFlow MVP is a lightweight AI market analysis script that drives GLM-4.7 model to output high-quality industry research briefs through carefully designed structured prompts. Target Users: Personal developers/analysts needing to quickly validate GLM-4.7's potential in research report generation. Scenario: Input industry keyword (like 'coffee chain'), directly generate Markdown report with market size and competitor analysis.
Debate Championship
Solves debate competition about chicken or egg first problem.
Family Travel Planner
Family members have different travel preferences.
Historical Traces
Historical Traces is a creation system transforming historical themes into professional scripts through Multi-Agent collaboration. Solves traditional script creation being time-consuming, requiring extensive historical research, lacking user engagement. In educational scenarios, students struggle to experience historical events immersively. Innovative Agent collaboration: 1. Linear + bidirectional hybrid: AgentA→B→C forms linear research chain, AgentC conducts multi-round bidirectional interaction with users, AgentD silently collects full-process info. 2. Central coordination: AgentG as sole user interface, unified info flow and task distribution management. 3. State machine interaction control: AgentC implements strict 'one question one answer' state management through 4-layer filtering, ensuring user deep participation in each creation stage.
TongTong
This project builds a multi-agent collaboration system based on OpenAgents, focusing on personal daily task management needs, integrating Mod-driven functions with multi-Agent division of labor, balancing technical depth, scenario innovation and application value, fully meeting advanced task criteria.
Three Kingdoms Advisors
A multi-agent dialogue system for student users, simulating seven famous advisors from the Three Kingdoms period (Zhuge Liang, Sima Yi, Guo Jia, Zhou Yu, Pang Tong, Xun Yu, Jia Xu), helping analyze various study, life and future planning choices from different perspectives.
Dimensional Companion - Honkai Study Partner
This project is a multi-agent emotional companionship system built on OpenAgents Enterprise framework, letting popular characters from the 'Honkai' series (like Mei, Kiana, Bronya, etc.) 'break through dimensional barriers' as AI agents, providing exclusive virtual companionship with long-term memory, emotional perception and personalized response.
Java Interview Chat Room
Solves limited interview opportunities, need for good preparation, need for mock interview practice.
Job Search Agent Suite
Job search assistant agent cluster developed with OpenAgents + Zhipu GLM-4-plus, covering full process from 'resume translation → resume optimization → interview coaching → salary calculation', adapting to various job positions, supporting Chinese/English job search scenarios, no concurrency limits, one-click start.
Intranet Knowledge Base Agent Accessible via Internet
Can penetrate intranet from public internet for knowledge base RAG and Text2SQL data queries.
OpenAgents Investment Committee Room
A multi-agent chat room where AI agents roleplay as startup team members discussing and debating startup ideas.
New Sales Scenario Dialogue Simulation Assistant
A multi-agent scenario dialogue where each agent plays different roles with different tones, for new salesperson (local life services as example) script practice and scenario simulation, improving new sales script expression and scenario response capabilities.
AI Crosstalk Troupe
This project builds a 'Workplace Fun Crosstalk Troupe' multi-agent system based on OpenAgents + Qianwen, with comedian, straight man, and coordinator three AI roles collaborating to auto-generate workplace-themed crosstalk. Core creativity focuses on worker high-frequency complaints, transforming AI office, useless overtime and other annoyances into dense laughs, using 'automated crosstalk pair' to replace manual scriptwriting, achieving complete performance on command trigger. System is beginner-friendly, 30-minute setup, satisfying stress relief entertainment and team building needs while providing concrete practice case for AI multi-agent technology, balancing fun and practicality.
Education Network
Many parents and teachers are recently discussing high school entrance exam (Zhongkao) matters. I applied 'multi-agent collaboration' technology to this scenario - building multiple subject teacher agents deeply engaged in Zhongkao, constructing group chat network through OpenAgents, using base models to give them professional personas and dialogue logic, having them discuss exam point changes, review methods, question trends, providing practical reference for students/parents while demonstrating AI Agent value in vertical scenarios.
Judging Panel
14 expert judges evaluating the hackathon projects
Lan Tian (蓝天)
Founder, Peak Mojo
Dedicated to building AI-era talent assessment infrastructure. Her team collaborates with researchers from Stanford, University of Minnesota, and Sun Yat-sen University to create industry standards for “Human-AI Collaboration Capability” and “AIQ (Artificial Intelligence Quotient)”. Before founding Peak Mojo, she was an early Uber engineer and worked as a venture capitalist, investing in and mentoring multiple AI startups.
Chang Gaowei (常高伟)
Founder, ANP Community
Founder of the Agent Network Protocol (ANP) open-source tech community, Co-Chair of W3C AI Agent Protocol CG, Founder of Hangzhou Vector Consensus, former Alibaba senior tech expert, and main author of ANP - the world’s first open-source agent communication protocol.
Lin Yuanlin (林沅霖)
Founder & CEO, Zeabur
CS graduate from Zhejiang University. In 2023, his undergraduate thesis project Zeabur was selected for Qiji Venture’s accelerator with $300K seed funding. Building an AI DevOps engineer that enables anyone to deploy, monitor, backup, and test services using natural language. Zeabur raised $2M angel round led by 500 Global in Fall 2025 and is expanding globally.
Zhang Xiaopeng (张晓鹏)
Head of Research, Guancha
Head of Research at Guancha & TegongYuzhou, leading Agent development and projects including Madu Lai, Image Arena, and EverythingRank. Guancha is the TapTap for AI agents - a content community and product distribution platform serving AI entrepreneurs, super individuals, investors, and users with a complete path from discovery to community building.
Zhang Zhuo (张卓)
Founder, AINow
Discovering the most important creators in AI. Former 36Kr VP and Renwu Magazine Deputy Editor. Journalism undergraduate from Renmin University, Marketing Master’s from HKU.
Si Jinqi (司晋琦)
W3C ANP Tech Advisor
20 years of product R&D at major tech companies. Currently W3C Agent Network Protocol Technical Advisor, focusing on full-stack AI. Former CTO of Shanda Literature Group (Qidian), VP of Tencent Yuewen Group, and CTO of PPDAI FinTech Group, with 15 years of strategic executive experience.
Iris (生姜)
Cofounder, AFFiNE
AFFiNE Cofounder & ex-COO (raised $10M total), Qiji S21 alumni. Edinburgh University PhD dropout turned entrepreneur, 4 years from seed to Pre-A. Forbes Asia 30U30. RTE community advisor, DKU startup mentor. 3 years overseas operations experience reaching 100+ countries. Mentored 30+ projects to Product Hunt #1. AFFiNE launch: 6K stars in one week, 10K in one month.
Steve Shao
Algorithm Lead, YouWare
AI algorithm engineer, formerly at TikTok and MiniMax leading algorithm optimization. Deep technical understanding and practical experience in AI Agents, LLMs, recommendation systems, and content search. Currently Algorithm and Agent Lead at YouWare, a Vibe Coding community providing intelligent Coding Agents where everyone can create, share, and explore multi-agent collaboration.
Li Yihao (李一豪)
Partner, Creekstone Ventures
8 years VC experience, focusing on native AI applications, agent infrastructure, and AI hardware. Providing deep service and long-term support for disruptive innovation, vision-driven founders with native tech understanding. Portfolio includes YouWare, Yanchuang Wanwu, Odyss, Daqian, MemU, and Mizzen.
Dreamy Li
Host, Simu Xiangduilun
An AI enthusiast who previously worked in investment research at a top Chinese VC covering AI and robotics. Former senior analyst at 36Kr tech sector, having engaged with 1000+ startups. Simu Xiangduilun is a tech media and community focusing on AI, robotics, and overseas expansion, covering interesting technologies and products that are changing the world.
Chen Hong (陈宏)
Founder, memU
AI entrepreneur and LLM researcher. Previously at Huawei, Japan AIST, TikTok, and ByteDance with years of experience in LLM optimization, dialogue systems, and AI research. Founder of MemU, focusing on Agent Memory. MemU provides open-source frameworks and enterprise services to give AI applications long-term memory - enabling agents to accumulate user knowledge, maintain context, and evolve across sessions.
Fan Wendong (范文栋)
Core Member, CAMEL AI
AI engineer and open-source contributor, currently Founding Engineer at Eigent AI and core member of CAMEL AI. Statistics Master’s with honors from University College Dublin, European Master in Official Statistics (EMOS) certified. Former AI engineer at BASF China Digital Center, building and deploying multiple industrial AI products. Part-time researcher at KAUST focusing on multi-agent framework development.
Gao Liming (高利明)
Investor & Founder, Stockstar
CEIBS EMBA graduate. Angel investor backing 70+ founders with multiple IPO exits. Passionate about supporting young entrepreneurs. Author of the “Bujie Nanti” (Unsolved Problems) public account.
Dalton
Investor, Monolith Management
Focused on early-to-mid stage AI & Deeptech investments. Monolith Management was founded in 2021 by former Sequoia China partner Cao Xi and former Boyu Capital partner Tim Wang. Previously at Planetree. Portfolio includes Datology.AI, Creao.AI, Knora.AI, PAML Biotechnologies. Former Cambridge Research Associate focusing on gradient descent, nonlinear non-convex optimization, and Bayesian optimization. CUHK Math & Economics degree.
Community Partners
Organizations supporting the Multi-Agent Hackathon 2025
Technology Ecosystem Partners
Technology partners powering the hackathon
Ready to Build Your Own Multi-Agent System?
Join the OpenAgents community and start building powerful AI agent networks today. Whether you are creating productivity tools, research assistants, or creative applications, OpenAgents provides the foundation you need.