Precis

Precis

本地优先的可视化数据质量工具 / Local-First Visual Data Quality Tool

可视化建模 · 全链路校验 · 数据不出本机

中文 · English

Beta — 核心功能稳定，V2 配置格式与 CLI/API 契约已冻结（只增不减）。生产关键数据建议保留备份。 欢迎通过 Issues、Discussions 与 Pull Request 参与贡献，贡献方式见 CONTRIBUTING.md。

Beta stage. Core features are stable; the V2 config format and CLI/API contracts are frozen (additive changes only). Keep backups for production-critical data. Contributions are welcome via Issues, Discussions, and Pull Requests — see CONTRIBUTING.md to get started.

---

项目简介

Precis 是一款针对 Excel / CSV / TSV 表格数据的质量校验工具。校验流程在可视化画布上以节点和连线的方式编排——从数据接入、清洗转换到多维度质量检查，全程无需编写代码。所有数据均在本地处理，不上传至任何外部服务。

提供三种使用方式，按需选择：

• 桌面应用（推荐）：图形界面，开箱即用
• 命令行：适合批量执行，或交由 AI 编程助手（Kimi Code / Claude Code 等）调用
• HTTP 接口：用于集成到自有系统

功能特性

• 可视化编排 — 在画布上拖拽节点、连接流程，即可完成校验建模
• AI 助手 — 接入国内外大模型或本地 Ollama，在对话中用自然语言推断 Schema、生成约束，确认后自动写入并同步画布；CLI 同样可用（ai chat / ai generate）
• 10 种检查规则 — 必填校验、唯一性、引用完整性、允许值清单、数值范围、条件判断、自定义脚本、字符集、日期逻辑、多规则组合
• 22 种数据转换 — 字符串拆分、模式提取、数学计算、分组聚合、过滤、排序等
• 大文件支持 — 超大文件自动分块处理
• 本地运行 — 数据不出本机；中英文双语界面；桌面应用内置运行环境，安装即可使用

快速开始

环境要求

如何确认本机版本

node --version          # 应 ≥ 20.19.0 或 ≥ 22.12.0
python3 --version       # 应为 3.12.x 或 3.13.x

若系统默认 python3 低于 3.12（macOS 自带 3.9），可通过 Homebrew 安装：

brew install python@3.12    # 安装后以 python3.12 调用

安装

git clone https://github.com/AirSaiga/Precis.git
cd Precis
npm run setup:mac              # macOS / Linux
# npm run setup:win            # Windows

一键脚本将依次完成：Python 环境检测、虚拟环境创建、依赖安装、前端与桌面应用构建。

无需图形界面、仅需在命令行或 AI 助手中执行校验，可直接 pip install precis-cli 获取 precis 命令；接入 AI 编程助手的方式见 integrations/。

开发者手动安装（含测试工具）

git clone https://github.com/AirSaiga/Precis.git
cd Precis

# 1. 安装前端依赖（根目录一次安装，覆盖全部子项目）
npm run install:all

# 2. 配置后端环境
cd backend
python3.12 -m venv .venv               # 若命令为 python3.13 则相应替换
source .venv/bin/activate              # Windows PowerShell: .venv\Scripts\Activate.ps1；Git Bash: source .venv/Scripts/activate
pip install --upgrade pip
pip install -e ".[dev]"                # 含 pytest / ruff / mypy 等开发工具
cd ..

注意：凡调用 python 的 npm 脚本（dev / backend:dev / cli / electron:dev），执行前须先运行 source backend/.venv/bin/activate，否则将使用系统自带的低版本 Python。

启动

npm run electron:dev          # 桌面应用（推荐）：自动启动后端与前端
npm run dev                   # 开发模式：前后端分离运行，支持热更新
npm run cli                   # 纯命令行，不启动图形界面
npm run start:tui             # 终端界面（实验性，不随发布版本提供）

更多启动方式与脚本说明见 scripts/README.md 与 tui-rust/README.md。

验证安装

npm run cli:validate          # 使用内置示例数据（qa_test/qa_simple/）执行一次校验

正常输出校验结果即表示环境就绪。

AI 助手（对话式建模）

除手动编排外，还可以让 AI 代劳。在应用设置中接入大模型——内置 OpenAI、Anthropic、Gemini、DeepSeek、通义千问、智谱、Kimi、MiniMax、小米 MiMo 等服务商预设，也支持本地 Ollama——之后即可在对话中描述需求：

• AI 推断数据结构，生成 Schema 与约束草稿
• 每次写盘前列出变更清单，确认后应用到项目并同步到画布
• 命令行同样可用：precis 交互式 shell 中先以 provider 配置服务商，再执行 ai chat（Agent 模式）或 ai generate data.xlsx 从数据文件直接生成配置

AI 功能为可选：不配置任何服务商时，校验与其他功能不受影响，全部校验仍在本地执行。为外部 AI 编程助手接入的方式见下一节 MCP Server。

MCP Server（AI 助手直连）

本仓库实现了一个 Model Context Protocol (MCP) server（stdio 传输，基于官方 MCP Python SDK）。支持 MCP 的 AI 编程助手（Claude Code、Cursor、Kimi Code 等）配置后可直接调用校验引擎。

pip install "precis-cli[mcp]"     # 安装后获得 precis-mcp 命令

在 MCP 客户端中配置（mcpServers 格式，Claude Code / Cursor 等通用）：

{
  "mcpServers": {
    "precis": {
      "command": "precis-mcp"
    }
  }
}

提供 4 个工具：

更多 AI 助手接入方式（skill、插件包等）见 integrations/。

项目结构

Precis/
├── backend/        # 校验引擎与命令行
├── frontend/       # 可视化编辑器
├── electron/       # 桌面应用外壳
├── tui-rust/       # 终端界面
├── e2e/            # 自动化测试
├── qa_test/        # 内置示例数据
├── scripts/        # 构建与部署脚本
└── docs/           # 架构文档

更多文档

• 开发者：开发命令与打包说明见 CONTRIBUTING.md；架构细节见 AGENTS.md 与 docs/ARCHITECTURE.md
• CHANGELOG.md — 变更日志
• SECURITY.md — 安全说明
• 各子项目详情：backend · frontend · electron · tui-rust · scripts · e2e

许可证

Apache-2.0 — 详见 LICENSE_NOTICE.md。

---

Overview

Precis is a data quality tool for Excel / CSV / TSV tabular data. Validation workflows are composed on a visual canvas using nodes and connections — from data ingestion and transformation to multi-dimensional quality checks, without writing any code. All data is processed locally and never uploaded to any external service.

Three usage modes, choose as needed:

• Desktop app (recommended): graphical interface, ready out of the box
• Command line: suitable for batch execution, or for invocation by AI coding assistants (Kimi Code / Claude Code, etc.)
• HTTP API: for integration into your own systems

Features

• Visual composition — build validation workflows by arranging nodes and connections on a canvas
• AI assistant — connect a cloud model or a local Ollama instance, then infer schemas and generate constraints from natural-language descriptions; confirmed changes are applied to the project and synced to the canvas; also available in the CLI (ai chat / ai generate)
• 10 check rules — required values, uniqueness, referential integrity, allowed value lists, numeric ranges, conditional checks, custom scripts, character sets, date logic, and multi-rule combinations
• 22 data transforms — string splitting, pattern extraction, math evaluation, grouping and aggregation, filtering, sorting, and more
• Large file support — oversized files are automatically processed in chunks
• Runs locally — data never leaves your machine; bilingual Chinese/English interface; the desktop app bundles its runtime and works immediately after installation

Quick Start

Prerequisites

How to check your local versions

node --version          # should be ≥ 20.19.0 or ≥ 22.12.0
python3 --version       # should be 3.12.x or 3.13.x

If the system default python3 is older than 3.12 (macOS ships 3.9), install one via Homebrew:

brew install python@3.12    # invoke as python3.12 after installation

Installation

git clone https://github.com/AirSaiga/Precis.git
cd Precis
npm run setup:mac              # macOS / Linux
# npm run setup:win            # Windows

The setup script performs, in order: Python detection, virtual environment creation, dependency installation, and frontend / desktop app build.

If you only need validation from the command line or an AI assistant, run pip install precis-cli to get the precis command; see integrations/ for AI coding assistant integration.

Manual install for developers (includes test tools)

git clone https://github.com/AirSaiga/Precis.git
cd Precis

# 1. Install frontend dependencies (single root install covering all sub-projects)
npm run install:all

# 2. Set up the backend environment
cd backend
python3.12 -m venv .venv               # substitute python3.13 if that is your command
source .venv/bin/activate              # Windows PowerShell: .venv\Scripts\Activate.ps1; Git Bash: source .venv/Scripts/activate
pip install --upgrade pip
pip install -e ".[dev]"                # includes pytest / ruff / mypy and other dev tools
cd ..

Note: any npm script that invokes python (dev / backend:dev / cli / electron:dev) requires source backend/.venv/bin/activate first; otherwise the system's older Python will be used.

Launch

npm run electron:dev          # Desktop app (recommended): starts backend and frontend automatically
npm run dev                   # Dev mode: backend and frontend run separately with hot reload
npm run cli                   # Command line only, no graphical interface
npm run start:tui             # Terminal UI (experimental, not included in releases)

Additional launch options and script details: scripts/README.md and tui-rust/README.md.

Verify the Installation

npm run cli:validate          # runs a validation pass on the bundled sample data (qa_test/qa_simple/)

Successful validation output indicates the environment is ready.

AI Assistant (Chat-Based Modeling)

Besides manual composition, you can let an AI do the wiring. Connect a model in the app settings — presets are bundled for OpenAI, Anthropic, Gemini, DeepSeek, Qwen, Zhipu GLM, Kimi, MiniMax, and Xiaomi MiMo, and a local Ollama instance is supported — then describe what you need in the in-app chat:

• The AI infers the data structure and drafts schemas and constraints
• Every write to disk is preceded by a change list; confirmed changes are applied to the project and synced to the canvas
• The CLI works the same way: configure a provider with provider inside the precis interactive shell, then run ai chat (agent mode) or ai generate data.xlsx to generate config straight from a data file

The AI features are optional: with no provider configured, validation and every other feature work as usual, and all validation still runs locally. To integrate external AI coding assistants instead, see the MCP Server section below.

MCP Server (for AI assistants)

This repository implements a Model Context Protocol (MCP) server over stdio, built on the official MCP Python SDK. MCP-capable AI coding assistants (Claude Code, Cursor, Kimi Code, etc.) can call the validation engine directly once configured.

pip install "precis-cli[mcp]"     # provides the precis-mcp command

Client configuration (mcpServers format, works with Claude Code / Cursor and others):

{
  "mcpServers": {
    "precis": {
      "command": "precis-mcp"
    }
  }
}

Four tools are exposed:

More AI assistant integration options (skills, plugin packages): integrations/.

Project Structure

Precis/
├── backend/        # Validation engine and command line
├── frontend/       # Visual editor
├── electron/       # Desktop app shell
├── tui-rust/       # Terminal interface
├── e2e/            # Automated tests
├── qa_test/        # Bundled sample data
├── scripts/        # Build and deployment scripts
└── docs/           # Architecture documentation

More Documentation

• Developers: dev commands and packaging notes in CONTRIBUTING.md; architecture details in AGENTS.md and docs/ARCHITECTURE.md
• CHANGELOG.md — Changelog
• SECURITY.md — Security notes
• Per-project details: backend · frontend · electron · tui-rust · scripts · e2e

License

Apache-2.0 — See LICENSE_NOTICE.md for details.