Clone Yourself — Upload your photo, text corpus, and voice sample to generate your digital twin.
| Feature | Description | Model |
|---|---|---|
| 🧠 Skill Cloning | Generate your thinking framework from text corpus | Nuwa-style LLM analysis |
| 🎤 Voice Cloning | Clone your voice from 1-10 min audio | GPT-SoVITS / CosyVoice2 / Fish Speech |
| 🖼️ 2D Avatar | One photo → animated talking video | LivePortrait / HunyuanVideo-Avatar |
| 🧊 3D Avatar | One photo → interactive 3D model | InstantMesh / TripoSR + Three.js |
| 🔬 Model Distillation | Compress large models for edge deployment | Knowledge Distillation pipeline |
- Local Mode — GPU inference, data never leaves your machine
- Cloud Mode — Third-party APIs, zero deployment
- Hybrid Mode — Core local + enhancement via cloud
| Layer | Technology |
|---|---|
| Backend | Go (Gin + pgx/pgxpool + Viper + Zap) |
| Frontend | React 18 + TypeScript + Three.js + Ant Design |
| Database | SQLite (modernc, pure-Go, single file) |
| Storage | MinIO (S3-compatible) |
| ML | PyTorch (called from Go via subprocess/gRPC) |
git clone https://github.com/Bruce-Sakura/UploadMyself.git
cd UploadMyself
bash install.sh # 交互选择 Docker / Linux 原生See INSTALL.md — supports Docker and Linux native, full features incl. GPU ML, runtime data under ~/.UploadMyself.
详见 INSTALL.md：支持 Docker 与 Linux 原生 两种方式，全功能含 GPU ML，运行时数据集中在 ~/.UploadMyself。
- Go 1.22+
- Node.js 18+
- Docker & Docker Compose
- NVIDIA GPU (for local ML inference)
git clone https://github.com/Bruce-Sakura/UploadMyself.git
cd UploadMyself
# Backend
cd backend && go mod tidy && cd ..
# Frontend
cd frontend && npm install && cd ..
# Download ML models
make models-download# SQLite 单文件数据库，无需额外依赖服务
# 推荐直接用一键安装: bash install.sh
# Backend (port 8000)
cd backend && go run .
# Frontend (port 5173)
cd frontend && npm run devmake build # Build Go binary
make frontend-build # Build React app
make test # Run all tests
make lint # Run linters
# Build the upme CLI
cd backend && go build -o upme ./cmd/upmeupme is a small CLI client for the backend REST API.
cd backend && go build -o upme ./cmd/upmeThe server address defaults to http://localhost:8000; override it with the
-server flag or the $UPME_SERVER environment variable.
upme health # health check
upme skill list # list all skills
upme skill get -id <id> # show one skill (incl. SKILL.md)
upme skill import -url <github/raw url> [-name <n>] # import from a URL
upme skill import -file <path> [-name <n>] # import from a local file
upme skill new -name <n> -corpus <text|@file> # generate from corpus (LLM)
upme skill rm -id <id> # delete a skill
upme chat -skill <id> -m "<message>" [-conv <id>] # chat with your twinBring a ready-made SKILL.md into the system without re-running the LLM:
POST /api/v1/skills/import
{ "url": "https://github.com/owner/repo/blob/main/path/SKILL.md", "name": "optional" }
# or
{ "content": "---\nname: my-skill\n---\n...", "name": "optional" }
- Provide either
urlorcontent. - GitHub
blobweb URLs are automatically rewritten to theirraw.githubusercontent.comform. - If
nameis omitted, it is taken from thename:field of the YAML frontmatter (the Claude Agent Skills format), falling back toimported-skill.
UploadMyself/
├── backend/ # Go backend (Gin, layered modules)
│ ├── main.go # Entry point: DI wiring, pgxpool, embedded migrations
│ ├── migrations/
│ │ └── 001_init.sql # DDL, embedded via go:embed & run at startup
│ ├── internal/
│ │ └── llm/client.go # OpenAI-compatible LLM client
│ ├── pkg/ # One module per table, six layers each
│ │ ├── tasks/ # ← module template (entity/dto/mapper/service/service/impl/handler)
│ │ │ ├── entity/ # table struct
│ │ │ ├── dto/ # request/response objects
│ │ │ ├── mapper/ # pgx data access
│ │ │ ├── service/ # business-logic interface
│ │ │ ├── service/impl/ # interface implementation
│ │ │ └── handler/ # HTTP handlers + route Register
│ │ ├── skills/ # Thinking-framework / SKILL.md generation
│ │ ├── voices/ # Voice cloning (train / synthesize)
│ │ ├── avatars/ # 2D/3D avatar processing
│ │ ├── file_uploads/ # File & corpus uploads (OCR/PDF/Word)
│ │ └── messages/ # Agent chat, conversation messages, tool registry
│ ├── cmd/
│ │ └── upme/ # `upme` CLI client for the REST API
│ └── middleware/ # CORS, etc.
├── ml/ # ML models & Python scripts
│ ├── models/ # Pretrained weights
│ └── scripts/ # Preprocessing & training (ml_service.py)
├── frontend/ # React + Three.js
│ └── src/
│ ├── pages/ # UI pages
│ └── three/ # 3D rendering
├── skills/ # Skill packages — <id>/SKILL.md (+ meta.json) per skill
├── docs/ # Documentation
└── tests/ # Tests
Skill storage — generated and imported thinking frameworks are stored as files, not blobs in the DB. Each skill is a directory
<SKILLS_DIR>/<id>/containingSKILL.md(the framework) andmeta.json(name/source/timestamp). The Postgresskillstable is only a metadata index (id, name, status…); theSKILL.mdbody is read from disk.SKILLS_DIRdefaults to./skillsand is mounted as a Docker volume so skills survive container restarts.
Module layering — every package under
backend/pkg/follows the same six-layer shape, withbackend/pkg/tasksas the reference template:entity(table struct) →dto(I/O objects) →mapper(pgx data access) →service(interface) →service/impl(implementation) →handler(HTTP + routes). Dependencies are wired inmain.goasmapper → service → handler. The legacybackend/agent,backend/handler, andbackend/modelpackages have been removed.
- Architecture
- API Reference — http://localhost:8000/docs after startup
- Deployment Guide
- Model Details
- Development TODO
Contributions welcome! See contributing.md.
MIT License — see LICENSE.
"The best way to predict the future is to create it." — Upload yourself into the digital world.
上传你的照片、文本语料、语音样本，生成你的数字分身。
| 功能 | 说明 | 模型方案 |
|---|---|---|
| 🧠 思维框架克隆 | 输入文本语料，生成你的思维模式 Skill | 仿女娲 + LLM |
| 🎤 语音克隆 | 1-10 分钟音频克隆你的声音 | GPT-SoVITS v2 / CosyVoice2 / Fish Speech v1.5 |
| 🖼️ 2D 虚拟形象 | 一张照片生成动态说话视频 | LivePortrait / HunyuanVideo-Avatar (2025最新) |
| 🧊 3D 虚拟形象 | 一张照片生成可交互 3D 模型 | InstantMesh / TripoSR + Three.js |
| 🔬 模型蒸馏 | 大模型压缩，降低部署成本 | NVIDIA TensorRT + 知识蒸馏 |
- 后端：Golang (Gin + pgx/pgxpool + Viper + Zap，分层模块结构)
- 前端：React 18 + TypeScript + Three.js + Ant Design
- 数据库：SQLite（modernc 纯 Go 驱动，单文件，零依赖）
- 存储：MinIO
- ML 推理：PyTorch (通过 Go subprocess/gRPC 调用)
git clone https://github.com/Bruce-Sakura/UploadMyself.git
cd UploadMyself
# 后端
cd backend && go mod tidy && cd ..
# 前端
cd frontend && npm install && cd ..
# 启动依赖
docker-compose up -d redis postgres minio
# 启动后端 (端口 8000)
cd backend && go run .
# 启动前端 (端口 5173)
cd frontend && npm run devupme 是后端 REST API 的命令行客户端。后端地址默认 http://localhost:8000，
可用 -server 或环境变量 $UPME_SERVER 覆盖。
# 构建
cd backend && go build -o upme ./cmd/upme
upme health # 健康检查
upme skill list # 列出所有思维框架
upme skill get -id <id> # 查看单个 skill（含 SKILL.md）
upme skill import -url <url> [-name <n>] # 从 URL/GitHub 导入
upme skill import -file <path> [-name <n>] # 从本地文件导入
upme skill new -name <n> -corpus <text|@file> # 用语料生成（触发 LLM）
upme skill rm -id <id> # 删除
upme chat -skill <id> -m "<消息>" [-conv <id>] # 与分身对话无需重新跑 LLM，即可导入现成的 SKILL.md：
POST /api/v1/skills/import
{ "url": "https://github.com/owner/repo/blob/main/path/SKILL.md", "name": "可选" }
# 或
{ "content": "---\nname: my-skill\n---\n...", "name": "可选" }
url与content二选一。- GitHub
blob网页地址会自动转换为raw.githubusercontent.com原始地址。 - 不传
name时，从 YAML frontmatter 的name:字段解析（Claude Agent Skills 格式）， 兜底为imported-skill。
Skill 文件存储：生成/导入的思维框架以文件形式保存在
<SKILLS_DIR>/<id>/SKILL.md（+meta.json），数据库skills表仅做元数据索引。SKILLS_DIR默认./skills，Docker 中挂载为skills卷。
详见 TODO.md — 包含 7 个阶段、80+ 个子任务的技术路线图。