DSH Data Agent · Analyze Data & Business Insights Through Conversation

English | English

         

专为数据分析与商业分析人打造的 DeepSeek 智能数据分析助手
自然语言提问 · 自动编写执行 SQL · 智能图表与报告 · AI 辅助业务治理 · 深度商业洞察 · 本地安全只读

产品亮点 · 快速上手 · 使用场景 · 工作台与报告 · 支持数据源 · 安全与隐私 · 常见问题 · 开源协议

Product Highlights

Every analysis means repeated SQL requests, manual Excel exports and merging, and format tweaks across chart tools? Complex tables and cryptic English abbreviations make business meanings impossible to look up?

DSH Data Agent brings data analysis back to its core: business decision-making:

• 💬 Zero-barrier conversational analysis: No more tedious SQL — just ask in plain language (e.g., "Compare conversion rates across channels in the last 30 days"). The AI understands business intent, explores schema, writes and runs queries, reasons step by step, and outputs clear conclusions.
• 📊 Auto high-quality charts & dashboards: No more dry text or black-and-white tables. From analysis conclusions it auto-draws line, bar, pie, scatter charts or multi-dimensional dashboards, with one-click export to offline HTML reports for sharing and reporting.
• 🧠 Deep business insights: Beyond listing numbers, it auto-compares trend fluctuations, pinpoints anomaly causes, and identifies high-value customer segments and best-selling categories, turning cold data into actionable business recommendations.
• 🏷️ AI-driven business metric governance: Scans the database and auto-annotates tables and fields with plain business meanings, supports human review and metric definition, ensuring every analysis is based on unified, accurate business semantics.
• 🔒 Local security & read-only protection: Natively supports read-only accounts and read-only mode; queries run locally under secure control, credentials stay strictly confidential, safeguarding enterprise production data.
• 🖥️ Modern web & efficient terminal, dual experience: Configure and browse charts in the visual web UI, or launch with one click from the efficient command-line terminal (dsh-tui).

Quick Start

1. Prepare the Runtime Environment

• DeepSeek Harness (DSH runtime environment)
• Local access to the target database (supports local SQLite files as well as LAN/cloud databases)

2. Quick Install

Run the following to install the plugin from npm:

# 安装到 Web 界面（推荐）
dsh plugin --profile web add @yejiming/dsh-data-agent

# 或安装到终端命令行（dsh-tui）
dsh plugin --profile dsh-tui add @yejiming/dsh-data-agent

3. Start Analyzing

On the first launch after upgrading the plugin, the recognized original data-agent preset is automatically updated to be compatible with the persona.config.prefix required by the new DSH, while the legacy text field used by older DSH versions is preserved. Customized presets will not be overwritten. If you encounter $.prefix missing required value on startup, back up $DSH_HOME/.agent-presets/data-agent/agent.cordis.yml (default location: ~/.dsh), make prefix in the persona's config use the same prompt as the original text, then restart.

Option 1: Web UI (recommended)

After launching the web console, create a new session and select "Data Mode":

dsh --profile web

1. Click "Connection config" next to "Data mode" on the new session page, or click the database icon at the top-right of the input box, then fill in connection info in the popup workspace (test connection available);
2. Once connected, type your business analysis question in natural language in the chat box;
3. Keep asking follow-ups; the AI drills down along the context layer by layer.

Option 2: Terminal command line (dsh-tui)

For keyboard-centric, geek users, connect in the terminal and start:

dsh --profile dsh-tui

In the session, type /preset data-agent to switch modes, then /database connect to connect to the database, and you're ready to start asking questions.

Use Cases

Scenario	Example Prompt
📈 Sales & Revenue Review	"Analyze sales and period-over-period changes across channels over the past 30 days, identify the category with the sharpest decline, and explain the main reasons"
👥 Customer Profiling & Segmentation	"Segment members by RFM based on purchase frequency and average order value over the past six months; compute retention rate and repurchase cycle per tier"
🛒 Conversion & Funnel Analysis	"For this month's new users, track conversion rates at each step — registration, search, add-to-cart, order — and identify the step with the highest drop-off"
📦 Inventory & Supply Chain Monitoring	"Find SKUs with turnover days exceeding 60, and forecast stockout or overstock risk based on recent sales velocity"
📑 Auto Weekly / Monthly Reports	"Summarize last week's key business metrics (GMV, active users, average order value) and generate an executive-level weekly report summary suitable for reporting"

Workspace & Report Outputs

1. Multi-function Database Workspace

The web UI includes a built-in, one-stop database workspace with four core modules — Connection Settings, Schema Browsing, Data Governance, and SQL Execution — so you can inspect your data assets any time besides chatting.

2. AI-assisted Data Governance

Open "Data Governance" in the workspace: the AI auto-scans table structures and generates clear Chinese business definitions, supporting item-by-item confirmation, deletion, and metric supplementation.

3. Interactive Offline Analysis Reports

When analysis requires visualization, the Agent automatically generates single charts or multi-dimensional dashboards and saves standalone offline HTML reports in the working directory's analysis-reports/ folder. The reports include rich metric cards, dynamic charts, and detailed data tables — open by double-click or send to colleagues anytime, no internet required.

Data Sources

DSH Data Agent broadly supports mainstream business databases, analytical data warehouses, and local files:

• 🐬 Relational databases: MySQL, PostgreSQL, SQLite, Oracle, Microsoft SQL Server
• ⚡ Analytical data warehouses / OLAP: ClickHouse, Apache Doris, Apache Hive, Apache Impala
• 📁 Local & lightweight data: SQLite data files (ready to use out of the box, no extra services required)

Security & Privacy

• 🛡️ Strict read-only protection: Use a read-only DB account and enable "read-only mode" on connect, ensuring no accidental modification or deletion.
• 🔑 Credential isolation: Passwords are used only during the current session, never logged in plaintext, and never sent to remote servers.
• 💻 Pure local execution: Queries and reports run locally under secure control, protecting enterprise business secrets.

FAQ

Q: 我完全不会写 SQL，能用它做好数据分析吗？

完全可以！DSH Data Agent 专门面向非技术背景的业务人员与分析师设计。你只需用平时的业务语言描述需求，AI 会自动理解意图、查找对应的数据表、编写并执行准确的 SQL 查询，并将结果整理成通俗易懂的商业图表与结论。

Q: 连接生产数据库会不会有误删或误改数据的风险？

不会。我们建议连接时配置数据库只读账号并勾选“只读模式”。在只读保护下，任何写入、修改、删除（如 UPDATE, DELETE, DROP）等高危指令都会被严格拦截，保障生产数据绝对安全。

Q: 公司的数据库表名和字段都是缩写拼音/英文，AI 能看懂吗？

可以。你可以使用内置的“数据治理”功能，让 AI 结合表注释和字段上下文一键扫描生成业务释义，你也可以手动录入公司特有的术语和计算公式（如“净成交额 = 订单金额 - 退款金额”），AI 会在后续分析中自动采用这些经过确认的业务口径。

Q: 生成的分析报告如何分享给不使用 DSH 的同事或领导？

每次生成的图表报告都会在本地保存为独立的 .html 文件（位于 analysis-reports/ 目录）。该文件内置了所有图表样式与交互数据，无需安装任何软件或连接网络，直接通过微信、邮件或钉钉发送给同事，用任意浏览器双击即可全屏查看与交互。

Q: 如果对分析出来的图表或结论有疑问，可以继续追问吗？

当然可以！就像和真实的专业分析师沟通一样，你可以在当前对话中继续追问（例如“再帮我按省份拆解一下”、“把柱状图换成饼图”、“为什么华东区 5 月份环比下降了”），AI 会基于上一步的结果继续深入分析。

License

This project is open-sourced under the MIT License.

Links

• dshfind.com: Discover DeepSeek Harness plugins and the broader ecosystem
• dsh-web-ui: Extensible web UI for DeepSeek Harness
• dsh-cc-tui: Full-screen interactive terminal UI for DeepSeek Harness
• platonai/Browser4: Large-scale web automation browser engine built for agents