GY

gradetwo/yanshi

开发工具

本页由 AI 翻译,技术细节请以原文为准。 查看原文

1 stars 质量 60 趋势 22MIT

- 每次编辑都是附加在日志上的原子,图片是日志的折叠。- 同一个Rust内核在服务器上原生运行,浏览器中作为WebAssembly运行,因此本地渲染和服务器......

概览

- 每次编辑都是附加在日志上的原子,图片是该日志的折叠。- 同一Rust内核在服务器上原生运行,浏览器中作为WebAssembly运行,因此本地渲染和服务器渲染是同一个实现,而非两种近似。- 一个 .yanshi 包携带 atom 日志及其引用的每个 blob,导入到其他地方时会重现图片。 MCP服务器和网页查看器都建在该引擎之上,所以代理和一个人操作同一个文档。 - — 由 rust-toolchain.toml 钉顶的稳定工具链,包括 rustfmt 和 clippy 组件。- 以及一个 POSIX shell — 以下每个入口点都是一个调用脚本的 Make 目标。- 可选,获取(浏览器内核):rustup target 添加 wasm32-unknown-unknown 和 wasm-bindgen-cli(cargo install wasm-bindgen-cli --version 0.2.129 --locked)。其版本必须与 wasm-bindgen crate 匹配,否则生成的胶水无法加载 .wasm。查看器在服务器端渲染。

概述(AI 翻译)

中文 |英语

Yanshi 是一个 AI 原生无头绘画引擎——一幅人与 AI 共同创作的画布。

  • 操作是原子。 每次编辑都是附加在日志上的原子,图片是日志的折叠。
  • 一个内核,两个主机。 同一Rust内核在服务器上原生运行,在浏览器中作为WebAssembly运行, 所以本地渲染和服务器渲染是一个实现,而不是两个近似。
  • 包是自包含的。 .yanshi 包携带原子日志及其引用的每个斑点,且 从其他地方导入就能重现画面。

MCP服务器和网页查看器都建在该引擎之上,所以代理和一个人操作同一个文档。

安装

预构建的二进制文件,不需要Rust工具链 — Linux x86_64(静态),Intel macOS,Apple Silicon。

这会解决你平台的发布资产,验证其SHA256与已发布的 SHA256SUMS,解包为~/.local/share/yanshi/ ,并置入yanshi-serve和yanshi-mcp 变成~/.local/bin,作为薄包装。

沥青球是自成一体的:画笔、画笔预览、媒介、调色板、材质素材、品牌资产

本页由 AI 翻译,技术细节请以原文为准。

README

以下为原文 README 节选,完整内容请查看原文。

Yanshi 偃师

中文 | English

Yanshi is an AI native headless painting engine — a canvas where people and AI create together.

  • Operations are atoms. Every edit is an atom appended to a log, and the picture is the fold of that log.
  • One kernel, two hosts. The same Rust kernel runs natively on the server and as WebAssembly in the browser, so local rendering and server rendering are one implementation, not two approximations.
  • A package is self-contained. A .yanshi package carries the atom log and every blob it references, and importing it elsewhere reproduces the picture.

An MCP server and a web viewer sit on top of that engine, so an agent and a person work on the same document.

Install

Prebuilt binaries, no Rust toolchain needed — Linux x86_64 (static), Intel macOS, Apple Silicon.

curl -fsSL https://raw.githubusercontent.com/gradetwo/yanshi/main/scripts/install.sh | sh

That resolves the release asset for your platform, verifies its SHA256 against the published SHA256SUMS, unpacks it to ~/.local/share/yanshi/, and puts yanshi-serve and yanshi-mcp into ~/.local/bin as thin wrappers.

Trust boundary: the checksum file ships in the same release as the tarball, so this check catches truncated downloads and mirror mix-ups — it does not protect against a compromised upstream release. Downloads are HTTPS-only (curl --proto/--proto-redir, wget --https-only).

The tarball is self-contained: brushes, brush previews, mediums, palettes, material assets, brand assets and the WASM kernel all ship inside it, and the binaries locate them through the package layout (bin/… next to share/yanshi/…), so nothing depends on where you unpack it.

yanshi-serve --root ~/paint --doc mydoc --bind 127.0.0.1:8080   # then open http://127.0.0.1:8080/
yanshi-mcp   --root ~/paint --doc mydoc                          # stdio MCP server, for MCP clients
yanshi-mcp   --list-tools --profiles core                        # just show the tool surface

Knobs: YANSHI_VERSION=v0.2.1 (default latest), YANSHI_PREFIX, YANSHI_BIN_DIR, and YANSHI_DRY_RUN=1 to print the plan without touching the disk.

Not covered by a prebuilt package: Linux arm64, and Windows. On those, build from source (cargo build --release -p yanshi-http -p yanshi-mcp, or make release) — or unpack the …-x86_64-pc-windows-msvc.tar.gz asset from the releases page by hand.

Quick start

Requirements

  • Rust — the stable toolchain pinned by rust-toolchain.toml, including the rustfmt and clippy components.
在 GitHub 查看完整 README

推荐工具

换一个关键词,或者移除筛选条件。

安装

npx skillfish add gradetwo/yanshi
新手?先看 mcpskill.org 教程 →