Datell

A polished local-first AI analyst for interactive reporting.

Upload files, connect databases, and describe the question in plain language. Datell turns that workflow into polished dashboards, export-ready reports, and presentations without forcing the model to reinvent the entire page every time.

上传文件、连接数据库、描述分析问题。Datell 会把这条工作流整理成交互式仪表盘、可导出的专业报表与演示文稿，而不是让模型每次从零开始“抽盲盒式”拼页面。

English · 中文

English

Datell is a polished local-first desktop analytics workspace built around a capable ReAct agent. It does more than answer questions: it can break work into steps, query databases, apply business context, and turn the result into interactive reports that feel ready to share.

Everything stays on-device by default, which makes Datell especially compelling when you want serious analysis, reusable outputs, and privacy without adding a cloud dependency.

At a Glance

Agentic analysis with visible reasoning, planning, execution, and verification
Interactive HTML reports with filter-aware KPI linkage, multiple chart engines, and export-ready layouts
Rich presentation surface with 170+ KPI cards, 40+ layouts, and built-in report presets
Extensible stack with multi-LLM support, RAG, knowledge graph, MCP tools, and external databases

Why Datell avoids prompt roulette

Many AI reporting products still rely on a fragile pattern: ask the model to improvise the page structure from scratch, hope the chart mix is reasonable, and hope filters, KPI summaries, and visual hierarchy happen to line up. That usually looks exciting in a demo and inconsistent in repeated use.

Datell takes a more systemized approach. The model works with a built-in report surface instead of an empty canvas:

20+ report presets that bundle chart engine, layout direction, and styling defaults
43+ layout templates spanning dashboards, documents, bento grids, wide screens, and poster formats
60+ palette presets with runtime palette injection for ECharts, ApexCharts, and VTable rendering
172 KPI and chart card patterns that give the agent stable building blocks instead of one-off HTML improvisation
Filter-aware interactivity runtime that coordinates controls, chart updates, KPI refresh paths, and DuckDB-backed data rebinding

That means the agent can still be creative where it matters, but the output is guided by reusable structure, visual consistency, and interaction contracts. In practice, Datell behaves much more like a report system with an AI operator than a prompt slot machine.

Download

Platform	File	Notes
Windows x64	`Datell-1.0.6-win-x64-portable.exe`	Portable, no install needed
macOS x64	`Datell-1.0.6-mac-x64.dmg`	Intel Mac
macOS arm64	`Datell-1.0.6-mac-arm64.dmg`	Apple Silicon (M1/M2/M3)
Linux x64 AppImage	`Datell-1.0.6.AppImage`	Portable AppImage for most distros
Linux x64 deb	`datell_1.0.6_amd64.deb`	Debian/Ubuntu package

→ View all releases

macOS first launch: If blocked by Gatekeeper, right-click the app → "Open".

Highlights

From data to report, in one message

Upload Excel / CSV or connect directly to a database. Describe what you want to analyze in plain language. Datell handles data cleaning, analysis, and visualization — outputting interactive HTML reports, Excel files, PDFs, or slide decks automatically.

Real agent, not just Q&A

Built on ReAct (Reasoning + Acting) architecture, the agent exposes its full reasoning process: Think → Plan → Execute → Verify. Every step is visible. Multi-step task progress panels make complex analyses transparent.

Professional reports, out of the box

Generated reports are fully interactive HTML dashboards — not static screenshots. Dynamic ECharts / ApexCharts, KPI cards with sparklines, filter linkage, DuckDB real-time SQL re-querying. Live preview on the right.

172 KPI card components

The built-in card library covers KPI metric cards (trend indicators, period-over-period), Sparkline line charts, mini bar charts, gauge dials, progress bars, comparison matrices, and more. The agent picks the right card type for your data automatically.

43+ layout templates across industries

Single-column, dual-pane dashboard, three-column wide, Bento Grid, magazine-wide, poster (portrait & landscape) — categorized by industry (Finance / Sales / HR / Marketing / Medical / Logistics). Switch the entire report layout with one click.

Real report example — monthly sales analysis

A complete monthly sales report auto-generated from sales data: 4 KPI cards, trend line chart, store-share pie chart, sales ranking bar chart, and a brand × product heatmap — all produced in a single pass.

Real report example — global energy data

A 60-year dataset covering 200+ countries. Datell automatically explores the data, computes statistics, compares trends, and generates a multi-dimensional visualization report.

Built-in SQLite User Databases — Large-Scale SQL Analysis

Create lightweight embedded SQLite databases directly inside Datell — no external DB server needed. The DB Management panel (Settings → DB Management) lets you create tables, browse schemas, import CSV/Excel data, and run arbitrary SQL queries against datasets of any size.

Real-scale test: 50,000-row sales dataset (2023–2024), 15 columns, 10 categories, 12 salespersons, 5 regions — imported in seconds, queried at sub-100ms per aggregation.

Table schema browser — 15-column sales_records table with rich field types:

SQL Step 1: Revenue breakdown by category — 10 rows, 42ms, aggregating across 50,000 records:

SQL Step 2: Monthly revenue trend — 24 rows (2 years), 58ms:

SQL Step 3: Salesperson performance ranking (executing — spinner active):

SQL Step 3 result: Top 10 salespersons, ordered by total revenue — 10 rows, 35ms:

Technical Architecture

Desktop shell: Electron 41 packages a TypeScript main process with a React 19 + Vite renderer, keeping the app portable across Windows, macOS, and Linux.
Secure data plane: database execution and credential handling stay in the main process, so the renderer works with sanitized results and orchestration APIs instead of raw secrets.
Report runtime: public/report-shell.html hosts preview and export rendering, applies layout and palette state, and keeps chart runtimes aligned across preview, export, and capture flows.
Interactivity layer: interactivity-engine.js and filter-controls.js coordinate filter changes, chart registration, KPI refresh paths, and DuckDB-backed rebinding for interactive reports.
Knowledge layer: SQLite persists app state locally, ONNX-based embeddings support local RAG, Kuzu provides the knowledge graph, and MCP endpoints extend the tool surface.

Features

AI Chat · ReAct Agent

Datell runs on the ReAct (Reasoning + Acting) architecture — the agent plans, calls tools, and iterates autonomously until the task is complete
Full conversation history in the sidebar with rename and search support
Multi-agent collaboration: parallel sub-agents, serial pipelines, aggregation nodes, nested calls
Real-time task progress panel for multi-step execution
ask_user tool for mid-task clarification (AG2UI interaction)
Thinking chain display, full support for reasoning models

Report Generation & Export

Output Format	Description
HTML Report	Interactive ECharts / ApexCharts, filter linkage, DuckDB dynamic SQL rebinding
VTable Big Data Table	Virtual scroll, 100K+ rows, pivot tables, tree tables, frozen columns
Excel File	Structured data export as `.xlsx`
PDF File	One-click HTML → PDF export
Slides / Presentation	Multi-page HTML slideshow with keyboard navigation, exportable to PDF
Document	Rich-text HTML document for printing or PDF export
Poster	Full-page single-card poster (portrait & landscape / 16:9)

Report Presets: 20+ bundled "palette + layout + chart engine" style packs (Business, Finance, Tech, Marketing, HR, Print)
Layout Template Library: 43+ visual grid layouts with custom editor
Color Palette Library: 60+ presets with custom editor
KPI Cards + Mini Charts (Sparkline, Gauge, Progress bar) embedded in reports

Multi-Database Connectivity

User SQLite Databases: create lightweight embedded SQLite databases directly inside the app — schema browsing, SQL console, CSV/Excel import, all offline
Supports MySQL / MariaDB / Apache Doris · PostgreSQL · Presto
Built-in SSH tunnel — no manual port forwarding needed
Connection pool management, automatic schema exploration (tables / column comments)
Natural language → SQL queries, results injected directly into reports

Knowledge Base (RAG)

Local vector store: ONNX local embedding model, fully offline
Dify: Connect to Dify API / external dataset retrieval
Ragflow: Connect to Ragflow document understanding pipeline
Custom chunking strategies (delimiter, max length, overlap)

Knowledge Graph

Built-in Kuzu graph database, persistent nodes and edges
Visual editor (add/delete nodes/relationships), agent can read/write the graph via tools

Model Compatibility

Provider	Notes
OpenAI	GPT-4o / GPT-4.1, multimodal vision
Anthropic	Claude 3.x / Claude 4 series
Google	Gemini 2.x Flash / Pro
Ollama	Local models (Qwen, Llama, Mistral, etc.)
OpenAI-compatible	Any compatible endpoint (SiliconFlow, DeepSeek, vLLM, etc.)
OpenRouter	Unified multi-model routing

Multiple model instances with free switching per chat; configurable baseUrl, maxTokens, temperature.

Tools & Extensions

Connect to any MCP (Model Context Protocol) HTTP/SSE server — tools are auto-discovered and registered
Custom skills: drop JSON skill files into datellData/skills/ for hot-loaded tools; the app currently scans *.json directory skills and can also install compatible skill.json payloads from URLs or GitHub repositories
Settings -> Skills now provides a unified panel for built-in tool manifests, registry-backed skills, legacy directory skills, and AI-created dynamic tools

Other

Dark / Light theme
Chinese / English UI (zh-CN / en-US)
File parsing: Excel (.xlsx), CSV, PDF, images (multimodal upload)
All data persisted locally via SQLite — nothing leaves your machine
Agent long-term memory, retains user preferences across sessions
web_fetch tool for fetching web content
Portable build for Windows, no installation needed

Getting Started

Prerequisites

Node.js ≥ 20.x
npm ≥ 10.x
Windows / macOS / Linux

Install Dependencies

git clone https://github.com/aiis2/datell.git
cd datell
npm install

Development Mode

npm run dev

This starts the Vite dev server (port 5173) and Electron main process concurrently. The system RAG index is built automatically on first run.

Building

# Windows portable build (x64)
npm run build:win

# macOS (x64 + arm64 DMG)
npm run build:mac

# Linux AppImage + deb
npm run build:linux

Build artifacts are output to the release/ directory.

Note: The release/ directory is excluded from git via .gitignore.

Configuration

Add AI Models

Go to Settings → Models to add model configurations:

Provider: OpenAI / Anthropic / Google / Ollama / OpenAI-compatible / OpenRouter
API Key: Enter the key for the corresponding service
Base URL: Set a custom address for proxies or compatible endpoints
Model ID: Model name (e.g. gpt-4o, claude-3-5-sonnet-20241022, gemini-2.0-flash)

Connect External Databases

Go to Settings → Datasources to add connections:

Supports MySQL / Apache Doris, PostgreSQL, Presto
Enable SSH tunnel and fill in bastion host details for automatic encrypted forwarding

Add Knowledge Bases

Go to Settings → Knowledge Base:

Local: Upload documents — the system auto-chunks, embeds locally, and indexes for full-text search
Dify: Enter API URL and Dataset ID
Ragflow: Enter API URL and Dataset ID

Custom Skills

Create JSON skill files in datellData/skills/ — the current loader scans *.json directory skills. Restart the app or use the refresh action in Settings -> Skills to reload them.

For file-backed custom skills with richer lifecycle management, use the registry section in Settings -> Skills:

Registry skills are stored under datellData/skills/registry/user/*.skill.json
You can create, edit, import, export, and delete registry skills from the UI
Legacy directory skills and AI-created dynamic tools can be promoted into the registry without changing the legacy compatibility loaders
Script-backed registry skills and dynamic skills can reuse enabled built-in tools via await callTool(name, args), which allows a custom skill to trigger generate_chart, generate_document, generate_slide, and other built-in report actions
A tracked example registry skill is available at skill/examples/visual-report-smoke.skill.json; it demonstrates composing HTML inside a custom skill and delegating the final preview render to callTool("generate_chart", ...)
A publish-ready Agent Skills working tree now lives under skill/publish/agentskills/; it currently exposes a single installable skill at skills/datell-visual-report-preview/ with an MCP-first workflow and a no-MCP fallback that still carries Datell card-library, layout, and palette guidance, while mcp/ remains the Track B visual-report runtime skeleton

License

This project is licensed under the Apache License 2.0.

Copyright 2026 Datell Contributors

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

中文

Datell 是一款打磨得相当完整的本地优先桌面数据分析工作台，核心是一套可执行、多步骤的 ReAct 智能体。它不只是回答问题，而是会主动拆解任务、查询数据库、结合业务上下文，并把结果整理成真正可交互、可分享的专业报表。

默认情况下，所有数据与对话都留在本机。这让 Datell 很适合既重视分析深度、结果复用，也重视隐私控制的团队和个人。

快速概览

可见的智能体分析流程：思考、规划、执行、验证完整展开
可交互 HTML 报表：筛选联动 KPI、图表与导出布局协同工作
很完整的展示层：170+ KPI 卡片、40+ 布局模板和多图表引擎
可扩展能力栈：多模型、RAG、知识图谱、MCP 工具与外部数据库

为什么 Datell 不会变成“报表抽盲盒”

很多 AI 报表产品仍然沿用一种很脆弱的方式：把整页结构交给模型临场发挥，期待它每次都能碰巧选对布局、图表组合、KPI 样式和筛选逻辑。演示时看起来很快，连续使用时往往就会暴露出不稳定和不一致。

Datell 走的是更系统化的路线。模型面对的不是一块完全空白的画布，而是一套内置的报表表达系统：

20+ 报告预设：把图表引擎、布局方向和样式默认值打包成稳定组合
43+ 布局模板：覆盖仪表盘、文档、Bento Grid、宽屏大屏和海报等不同场景
60+ 调色板预设：运行时可同步注入到 ECharts、ApexCharts 和 VTable
172 个 KPI / 图表卡片模式：让 Agent 基于稳定组件拼装，而不是每次即兴写一套 HTML
筛选联动运行时：统一处理控件、图表更新、KPI 刷新路径和基于 DuckDB 的数据重绑

这意味着 Agent 仍然可以在分析与表达上保持灵活，但输出结果会被可复用的结构、视觉一致性和交互契约约束住。实际体验上，Datell 更像是一套由 AI 驱动的专业报表系统，而不是一次次碰运气的提示词老虎机。

下载

平台	文件	说明
Windows x64	`Datell-1.0.6-win-x64-portable.exe`	免安装，双击直接运行
macOS x64	`Datell-1.0.6-mac-x64.dmg`	Intel Mac
macOS arm64	`Datell-1.0.6-mac-arm64.dmg`	Apple Silicon (M1/M2/M3)
Linux x64 AppImage	`Datell-1.0.6.AppImage`	适用于大多数发行版的便携 AppImage
Linux x64 deb	`datell_1.0.6_amd64.deb`	适用于 Debian / Ubuntu 的安装包

→ 查看所有发行版

macOS 首次打开说明：如提示「无法打开」，请右键点击应用 → 选择「打开」即可。

核心亮点

一句话，从数据到报表

上传 Excel / CSV，或直接连接数据库，用自然语言描述你要分析的内容，Agent 会自动完成数据清洗、分析、可视化，输出带交互图表的 HTML 报表、Excel 文件、PDF 或演示文稿。

真正的 Agent，不是简单的问答

基于 ReAct（Reasoning + Acting） 架构，Agent 会展示完整的推理过程：思考 → 规划 → 执行 → 验证，每一步都透明可见。支持多步骤任务进度面板，复杂分析任务不再是黑盒。

专业级报表，开箱即用

生成的报表不是静态截图，而是完全可交互的 HTML 大屏：ECharts / ApexCharts 动态图表、KPI 卡片带迷你趋势图、筛选器联动、DuckDB 实时 SQL 重查询。右侧实时预览，所见即所得。

精美的 KPI 卡片 + 迷你图表

内置 172 个卡片组件：KPI 指标卡（含趋势方向、环比变化）、Sparkline 折线图、迷你柱状图、仪表盘、进度条、对比矩阵……Agent 会根据数据自动选配最合适的卡片类型。

43 套布局模板，覆盖各类业务场景

单列文档、双栏仪表盘、三栏宽屏、Bento Grid、杂志宽版……按行业分类（Finance / Sales / HR / Marketing / Medical / Logistics），一键切换报表布局，无需手动排版。

真实报表示例 — 销售数据月报

以下是 Agent 基于销售数据自动生成的完整月报，包含 4 个 KPI 卡片、趋势折线图、门店饼图、业绩排名柱状图和品牌×商品热力图，全部由 AI 一次性自动完成。

真实报表示例 — 全球能源数据分析

跨越 60 年、覆盖 200+ 个国家的能源数据集，Agent 自动完成数据探索、统计汇总、趋势对比与分布分析，生成多维可视化报表。

内置 SQLite 用户数据库 — 大规模 SQL 分析

在 Datell 内部直接创建轻量级嵌入式 SQLite 数据库，无需任何外部数据库服务器。通过设置 → DB 管理面板，可创建表、浏览表结构、导入 CSV/Excel 数据，并执行任意 SQL 查询。

真实规模测试：50,000 条销售记录（2023–2024 年），15 列、10 个品类、12 名销售员、5 个大区，导入数秒完成，每次聚合查询响应时间低于 100ms。

表结构浏览器 — 15 列 sales_records 表，多种字段类型：

SQL 第 1 步：按品类汇总收入 — 10 行结果，对 5 万条记录聚合，耗时 42ms：

SQL 第 2 步：按月收入趋势 — 24 行（跨 2 年），58ms：

SQL 第 3 步：销售人员业绩排名（执行中 — 旋转动画）：

SQL 第 3 步结果：前 10 名销售人员按收入排序 — 10 行，35ms：

技术架构

桌面外壳：基于 Electron 41，采用 TypeScript 主进程和 React 19 + Vite 渲染层，统一打包 Windows、macOS、Linux。
安全数据面：数据库连接与查询执行留在主进程，渲染层拿到的是经过整理的结果和编排 API，而不是原始凭据。
报表运行时：public/report-shell.html 负责承载预览与导出渲染，统一应用布局、调色板和运行时注入逻辑，保证预览、导出、截图链路尽量一致。
交互层：interactivity-engine.js 与 filter-controls.js 负责筛选变化、图表注册、KPI 刷新路径以及基于 DuckDB 的联动重查询。
知识层：本地 SQLite 持久化应用状态，ONNX Embedding 支撑本地 RAG，Kuzu 提供知识图谱，MCP 负责外部工具扩展。

功能特性

AI 对话 · ReAct 智能体

基于 ReAct（Reasoning + Acting） 架构，Agent 自主规划、调用工具、循环推理直至任务完成
多轮对话历史，侧边栏支持重命名与搜索
多智能体协作：并行子 Agent、串行流水线、汇聚节点，支持嵌套调用
任务进度面板实时展示多步骤执行状态
ask_user 工具支持执行过程中向用户提问（AG2UI 交互）
思维链（Thinking）展示，支持推理模型

报表生成与导出

输出格式	描述
HTML 报表	含 ECharts / ApexCharts 交互图表，支持滤镜联动、DuckDB 动态 SQL 重绑
VTable 大数据表格	虚拟滚动，支持 10 万+ 行，透视表、树形表、冻结列
Excel 文件	结构化数据导出为 `.xlsx`
PDF 文件	HTML 报表一键导出 PDF（页面截图合并）
幻灯片/演示文稿	HTML 多页幻灯片，支持键盘翻页，可导出 PDF
专业文档	HTML 富文本文档，适合打印或导出 PDF

报告预设：内置 20+ 套"配色 + 布局 + 图表引擎"一键风格包（商业、财务、科技、营销、HR、打印等分类）
布局模板库：40+ 套可视化网格布局，支持自定义编辑
配色方案库：60+ 套调色板预设，支持自定义创建与编辑
KPI 卡片 + 迷你图表（Sparkline、仪表盘、进度条等）内嵌于报表

多数据库连接

用户 SQLite 数据库：在应用内直接创建轻量嵌入式 SQLite 数据库，支持表结构浏览、SQL 控制台、CSV/Excel 导入，完全离线
支持 MySQL / MariaDB / Apache Doris · PostgreSQL · Presto
内置 SSH 隧道，无需手动端口转发
连接池管理、Schema 自动探索（表结构 / 字段注释）
自然语言 → SQL 查询，结果实时注入报表

知识库（RAG）

本地向量库：基于 ONNX 本地 Embedding 模型，离线可用
Dify 知识库：接入 Dify API / 外部数据集检索
Ragflow 知识库：接入 Ragflow 文档理解流水线
自定义分块策略（分隔符、最大长度、重叠率）

知识图谱

内置 Kuzu 图数据库，持久化节点与边
可视化编辑器（增删节点/关系），Agent 可通过工具写入/查询图谱

模型兼容

提供商	说明
OpenAI	GPT-4o / GPT-4.1 等，含视觉多模态
Anthropic	Claude 3.x / Claude 4 系列
Google	Gemini 2.x Flash / Pro
Ollama	本地模型（Qwen、Llama、Mistral 等）
OpenAI 兼容	任意兼容接口（硅基流动、DeepSeek、vLLM 等）
OpenRouter	统一路由多模型

支持配置多个模型实例、聊天时自由切换；支持自定义 baseUrl、maxTokens、temperature。

工具与扩展

接入任意 MCP（Model Context Protocol）HTTP/SSE 服务器，自动发现并注册工具
自定义技能：在 datellData/skills/ 放置 JSON 技能文件即可加载目录技能；当前加载器会扫描 *.json 文件，也支持从 URL 或 GitHub 仓库安装兼容的 skill.json 内容
“设置 -> 技能” 已统一展示内置工具 manifest、registry 技能、legacy 目录技能与 AI 动态技能

其他

深色 / 浅色主题
中英文界面（zh-CN / en-US）
文件解析：Excel（.xlsx）、CSV、PDF、图片（多模态上传）
全量本地 SQLite 持久化，数据不离机
Agent 长期记忆，跨会话保留用户偏好
web_fetch 工具支持抓取网页内容
Windows 绿色便携版，无需安装

快速开始

前置条件

Node.js ≥ 20.x
npm ≥ 10.x
Windows / macOS / Linux

安装依赖

git clone https://github.com/aiis2/datell.git
cd datell
npm install

开发模式

npm run dev

启动后会并行运行 Vite 开发服务器（端口 5173）与 Electron 主进程。首次运行会自动构建系统 RAG 索引。

构建打包

# Windows 便携版（x64）
npm run build:win

# macOS（x64 + arm64 DMG）
npm run build:mac

# Linux AppImage + deb
npm run build:linux

构建产物输出至 release/ 目录。

注意：release/ 目录已在 .gitignore 中排除，不会提交到仓库。

配置说明

添加 AI 模型

在 设置 → 模型 中新增模型配置：

提供商：OpenAI / Anthropic / Google / Ollama / OpenAI 兼容 / OpenRouter
API Key：填入对应服务的密钥
Base URL：如使用代理或兼容接口，填入自定义地址
Model ID：模型名称（如 gpt-4o、claude-3-5-sonnet-20241022、gemini-2.0-flash）

连接外部数据库

在 设置 → 数据源 中添加数据库连接：

支持 MySQL / Apache Doris、PostgreSQL、Presto
可开启 SSH 隧道，填入跳板机信息后自动建立加密转发

添加知识库

在 设置 → 知识库 中：

本地知识库：上传文档，系统自动分块、本地 Embedding、全文检索
Dify：填入 API URL 和 Dataset ID
Ragflow：填入 API URL 和 Dataset ID

自定义技能

在 datellData/skills/ 目录创建 JSON 技能文件；当前加载器会扫描 *.json 目录技能。重启应用或在“设置 -> 技能”中点击刷新后即可重新加载。

如果需要更完整的技能生命周期管理，可直接使用“设置 -> 技能”中的 registry 区域：

registry 技能保存于 datellData/skills/registry/user/*.skill.json
支持在 UI 中新建、编辑、导入、导出、删除 registry 技能
也支持把 legacy 目录技能和 AI 动态技能提升到 registry，而不破坏现有兼容加载链路
脚本型 registry 技能和动态技能现在可以通过 await callTool(name, args) 复用已启用的内置工具，因此可直接在技能内部触发 generate_chart、generate_document、generate_slide 等报表动作并把结果送入预览面板
仓库已提供可追踪的真实示例技能 skill/examples/visual-report-smoke.skill.json，演示如何在技能内部组织 HTML，再通过 callTool("generate_chart", ...) 把可视化报表送入预览链路

许可证

本项目以 Apache License 2.0 开源。

Copyright 2026 Datell Contributors

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Name		Name	Last commit message	Last commit date
Latest commit History 41 Commits
.claude/skills		.claude/skills
.github		.github
.vite/deps		.vite/deps
build/icons		build/icons
docs		docs
plugins		plugins
public		public
resources/system_knowledge		resources/system_knowledge
scripts		scripts
skill		skill
src		src
tests		tests
.gitignore		.gitignore
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
build-win.cmd		build-win.cmd
index.html		index.html
package-lock.json		package-lock.json
package.json		package.json
postcss.config.js		postcss.config.js
skills-lock.json		skills-lock.json
tailwind.config.js		tailwind.config.js
tsconfig.json		tsconfig.json
vite.config.ts		vite.config.ts

Folders and files

Latest commit

History

Repository files navigation

Datell

English

At a Glance

Why Datell avoids prompt roulette

Download

Highlights

From data to report, in one message

Real agent, not just Q&A

Professional reports, out of the box

172 KPI card components

43+ layout templates across industries

Real report example — monthly sales analysis

Real report example — global energy data

Built-in SQLite User Databases — Large-Scale SQL Analysis

Technical Architecture

Features

AI Chat · ReAct Agent

Report Generation & Export

Multi-Database Connectivity

Knowledge Base (RAG)

Knowledge Graph

Model Compatibility

Tools & Extensions

Other

Getting Started

Prerequisites

Install Dependencies

Development Mode

Building

Configuration

Add AI Models

Connect External Databases

Add Knowledge Bases

Custom Skills

License

中文

快速概览

为什么 Datell 不会变成“报表抽盲盒”

下载

核心亮点

一句话，从数据到报表

真正的 Agent，不是简单的问答

专业级报表，开箱即用

精美的 KPI 卡片 + 迷你图表

43 套布局模板，覆盖各类业务场景

真实报表示例 — 销售数据月报

真实报表示例 — 全球能源数据分析

内置 SQLite 用户数据库 — 大规模 SQL 分析

技术架构

功能特性

AI 对话 · ReAct 智能体

报表生成与导出

多数据库连接

知识库（RAG）

知识图谱

模型兼容

工具与扩展

其他

快速开始

前置条件

安装依赖

开发模式

构建打包

配置说明

添加 AI 模型

连接外部数据库

添加知识库

自定义技能

许可证

About

Resources

License

Contributing

Security policy

Uh oh!

Stars

Packages