- HTML 53.2%
- JavaScript 29.4%
- TypeScript 17.4%
Reorder bilingual headings, paragraphs, table labels, video captions and navigation to English first. Lead with the actual English encounter screenshot and keep the full Chinese version and Chinese preview linked. |
||
|---|---|---|
| api | ||
| assets | ||
| docs | ||
| frontend | ||
| server | ||
| .env.example | ||
| .gitignore | ||
| AGENTS.md | ||
| package-lock.json | ||
| package.json | ||
| README.md | ||
| README.zh-CN.md | ||
| render.yaml | ||
| vercel.json | ||

Yorimi
Original characters. Ongoing connections. Across borders.
让原创角色走向世界,让关系持续发生。
AI Character Engine + a bilingual companion-game experience for Chinese original IP.
中国原创 IP 出海的 AI 角色引擎与双语陪伴游戏原型。
Watch film / 观看短片 · Play / 体验作品 · Agent Demo / AI 角色 · 简体中文 · Q2 Submission / 参赛材料

▶ Watch the film / 观看短片
29-second Chinese film with English subtitles · Seedance / CapCut. Character visits, shared worlds and hardware scenes illustrate development plans.
中文短片,配有英文字幕 · Seedance / CapCut · 29 秒。串门、共同世界与硬件画面为未来构想。

Actual public work, captured 2026-10-11. AI-assisted, authored branching dialogue; the character Agent is demonstrated separately.
2026-10-11 真实公开作品截图;AI 辅助创作的预写分支体验,角色 Agent 另行演示。
Full Chinese version / 完整中文版 · Chinese interface screenshot / 中文界面截图
The problem and the experience / 创作问题与体验
Chinese original game characters, independent IP, VTubers and OC creators need more than an initial overseas release: they need character-led interactions that remain recognisable and natural in the player's language. Yorimi explores that continuing relationship, starting with English-speaking and bilingual players in Singapore and Southeast Asia.
中国原创游戏角色、独立 IP、VTuber 与 OC 创作者进入海外市场后,需要持续制作符合角色身份、当地语言和玩家状态的内容。Yorimi 聚焦“IP 被看见之后,如何继续与海外玩家建立关系”,首批范围为新加坡与东南亚的英语、中英双语玩家。
One AI Character Engine lets a character respond to the player's latest message, retain its identity and continue a shared moment. Our companion-game prototype uses a playable loop: meet a character → choose how to participate → receive a response → follow up. The current cultural work turns tea and hospitality into a small invitation: “I saved you a seat / 给你留个位置.”
通过同一套 AI Character Engine,让角色根据玩家最新消息、身份设定和近期上下文继续回应。 当前循环:认识角色 → 选择参与方式 → 得到回应 → 继续互动。本轮作品用茶与待客表达“给你留个位置”。
Two deliverables, one cultural-export project / 两类参赛交付
| Direction / 提交方向 | What to inspect / 检查内容 | Entry / 入口 |
|---|---|---|
| Technical solution / Agent 技术方案 | Character context, executable skill selection, model generation, output guards, same-session continuity and machine-readable Agent Run | AI companion preview · Specs · Demo and evaluation |
| Creative work / AIGC 内容作品 | A Seat at the Tea Table: three player paths, three endings, English/Chinese switching, replay and original visual presentation | Play the complete work · Full bilingual text and actual AI process |
Entry / 参赛活动: Activity 1111 · Q2 Cultural Export, AI Game subtrack. Registered title: Yorimi|面向中国原创 IP 出海的 AI 角色陪伴养成游戏原型. Synnovator showed Approved for mart/Yorimi on 2026-10-11.
本项目提交 AI 角色引擎、技术 Specs 与评估依据,以及一部有完整选择和结尾的双语文化互动作品。技术与内容的实际证据分别检查。
29-second concept film / 29 秒概念短片 · Direct MP4 · Video, tools and production record
Review in three minutes / 三分钟评审路线
- Play the work / 先体验完整作品. Choose sharing, quiet company or a brief visit. Reach an ending, replay and switch language. The work runs publicly without an API key.
- Inspect the Agent / 再检查 Agent 执行. Follow the runtime demo; inspect intent, selected skills, outcomes, runtime source and checks in Agent Run. Use the local repository build for its current technical contract; the Vercel link is the existing companion preview.
- Check the creation evidence / 核对创作过程. Open the work/process record, creator localization pack and rights register.
Evidence for every scoring criterion / 八项评分证据
The official rules allocate 50% to the technical solution and 50% to the creative work. These links map the entry to the criteria; they are evidence, not claimed scores.
技术方案与内容成果各占 50%。下面逐项链接到实际材料,不预填官方得分;效率、文化/语言人工评分只在真实测量后记录。
| Criterion / 评分项 | Weight / 权重 | Yorimi evidence / 证据 |
|---|---|---|
| Needs and scenario / 需求与场景 | 10% | Specs: creators, overseas players, target market and recurring content problem |
| Agent design / Agent 方案 | 20% | Architecture, task breakdown, tools and human review · executable Skill Registry |
| Creative enablement / 创作赋能 | 10% | Reusable character brief, bilingual decisions and review handoff · time/quality comparison protocol |
| Feasibility and innovation / 可行性与创新 | 10% | Runnable build and six contract suites · same-engine reuse plan |
| Content quality and completeness / 内容质量与完成度 | 20% | Complete playable work · all dialogue, choices and endings |
| Cultural expression and adaptation / 文化表达与适配 | 15% | Tea/hospitality context, natural English and player choice · culture and language evaluation |
| Creativity and communication potential / 创意与传播潜力 | 10% | Three ways to share the same welcome · demo and future-concept storyboard |
| Originality and compliance / 原创性与合规 | 5% | Actual AI tools, revisions and material provenance · team rights review |
How the Agent works / Agent 如何工作
Player message + character profile + recent session context
→ infer intent and mode → select executable skill plan
→ provide the plan to the model → generate a structured response
→ normalize and check → render the answer and optional action
→ update the bounded session → expose Agent Run → player follow-up
One orchestrating Agent selects from eight executable skills: persona_adaptation, memory_continuity, concept_guidance, task_planning, emotional_support, quiz_practice, routine_progress and companion_presence. The model generates the natural-language reply; local code manages context, routing, schema, guards, session memory and evidence.
Inspect the Registry, orchestrator, OpenAI adapter, HTTP handlers and session store. GET /api/skills exposes the public catalog; successful POST /api/chat responses include agent_run. Model and fallback status are visible in each run. Optional Fish Audio and browser speech enhance voice interaction.
一个编排 Agent 在生成前选择可执行 Skill plan,并将角色、当前消息与近期上下文交给模型;本地代码负责 schema、输出检查、有界会话与 Agent Run。每轮显示真实运行来源。
What the cultural work expresses / 文化作品与双语适配
Luma is a fictional desk-lamp companion; Yori welcomes the player into the story as the Yorimi brand character. The original tea-table scene gives sharing, silence and leaving equal room. Each choice changes the ending.
English is adapted by intent: “给你留个位置” becomes “I saved you a seat,” while “安静坐一会儿” becomes “Sit here quietly.” Tea/hospitality is a cultural reference with local and family variation. The source and adaptation notes distinguish cultural background from fictional dialogue.
Actual AI use: Codex / ChatGPT assisted the text, SVG/CSS and code in this development session. The teammate-produced concept film uses Seedance and CapCut among its tools, as confirmed by Martina; its actual delivery and web conversion are recorded separately. Original YorimiLab artwork and the existing character animation are reused with their provenance recorded. The public deployment record covers the eight-file bundle, delivered film and verified public assets.
Luma 是虚构桌灯伙伴,Yori 是品牌引导形象。玩家可以分享、安静或离开;每一种选择都有收尾。英语按交流意图改写,文化解释承认地方和家庭习惯差异。实际 Codex / ChatGPT 创作过程、素材来源和人工复核在材料中分别记录。
Team, rights and history / 团队、权利与沿革
Wang Siyu(王斯昱) — Co-founder, Marketing & Business. ESSEC Business School Global BBA student and ESSEC Excellence Scholarship recipient, with AI consulting and generative AI evaluation experience at Mercor and market research/data analysis at Yunnan Yunda Coffee. Leads market positioning, creator/IP partnerships and commercial strategy.
Martina Lee — Co-founder, Product, AI Engineering & Hardware Integration. Physics PhD student at Nanyang Technological University, working across Physics and AI. Leads LLM character interaction, Agent orchestration/evaluation and AI product implementation, alongside hardware-integration feasibility.
王斯昱就读 ESSEC Global BBA,负责市场、创作者/IP 合作与商业策略;Martina 为 NTU Physics PhD 在读,负责 LLM 角色交互、Agent 编排与评估及产品实现。完整履历与分工见下方链接。
Full bilingual profiles and responsibilities. Yorimi evolved from Compagnon Éveil; the predecessor team was a UCWS Singapore Hackathon 2026 Top 20 Finalist (certificate).
The current build provides character interaction, bounded process-local session memory and a complete authored vignette. Creator Studio, user-governed persistent memory, relationship progression, character visits and multiplayer are next-stage plans. Future-feature shots in the storyboard carry a concept label. Human cultural review, efficiency measurements and final asset-rights confirmation are tracked in the submission guide.
当前已交付角色交互、有界进程内会话记忆和完整互动小品;Creator Studio、持久记忆、关系成长、串门和多人属于后续规划。概念视频标注未来画面;人工文化评价、效率实测与最终素材权利确认按实际完成状态追踪。
Reproduce the repository build / 复现仓库版本
Git and Node.js 18+:
通过 Synnovator 克隆,使用 Git 与 Node.js 18+。模型 key 放在不跟踪的 .env,语音为可选增强。
git clone https://www.synnovator.com/mart/Yorimi.git
cd Yorimi
npm ci
Copy-Item .env.example .env
npm run dev
Open http://127.0.0.1:3017/yorimi.html for the Agent and http://127.0.0.1:3017/culture-demo.html for the work. Configure OPENAI_API_KEY in the untracked .env and USE_MOCK_AI=false for model generation; Fish Audio is optional. The main companion frontend uses CDN dependencies; the cultural work is self-contained.
npm run test:all
# With OPENAI_API_KEY configured:
npm run test:openai
Q2 submission guide / 参赛材料 · Requirements map / 完整要求与评分证据 · Verification / 工程验证 · Registration / 报名记录