Introduction

Developing agents within DeepSeek Harness (DSH) often involves data analysis: business users want to ask questions like “Which region had the biggest sales decline last month?” directly, but engineers have to repeatedly switch to BI tools, manually write SQL, and then paste the results back into the conversation. DSH’s core philosophy is “everything is a plugin”—community members have already turned “connecting to databases for data querying” into a standalone plugin.

Today, we’re introducing dsh-data-agent (repository omdsh-dev/dsh-data-agent). It has gained 164 stars and 12 forks on GitHub and is categorized as a “networked tool” in the SkillHub Plugin Directory. Maintained by the community organization omdsh-dev under the MIT license, it’s important to note that SkillHub and directories like dshfind are community-run DSH plugin indexes, not official DeepSeek or High-Flyer app stores.

What Is It

In one sentence: dsh-data-agent is DSH’s data analysis plugin that integrates database connections, SQL execution, and result interpretation into a single conversational workflow.

After connecting to a database, you can describe business questions in natural language. DSH will examine the database schema, write and execute SQL, then continue to adjust queries or perform deeper analysis based on actual results, ultimately providing readable data conclusions and business insights. The plugin supports both Web UI and dsh-tui without modifying DSH source code; the npm package name is @yejiming/dsh-data-agent.

Core Features and Highlights

Based on the GitHub README and repository documentation, the key verified capabilities are as follows:

  1. Conversational Data Analysis: Ask questions in natural language, and DSH will break down analysis steps, query real data, and organize conclusions; supports follow-up questions in the same context to gradually narrow scope or compare dimensions.
  2. Automated SQL Closed Loop: Instead of just generating unverified SQL, it writes queries based on table structures, executes them, and iterates based on errors or results until a usable answer is obtained.
  3. AI-Driven Data Governance (Catalog): The model configured in the current session can scan databases to generate business meaning candidates for tables and fields; all AI suggestions require manual confirmation, and terminology and metric definitions can also be added manually. Subsequent analysis reads confirmed definitions through tools like catalog-search, catalog-get, and metric-get.
  4. HTML Analysis Report (render-analysis): When the Agent determines visualization is needed, it can generate a comprehensive report containing metrics, line charts, bar charts, pie charts, scatter plots, tables, and other views in one call, saving an offline HTML file to the analysis-reports/ directory; Web UI supports inline preview and “View Analysis” pop-ups.
  5. Unified Tool Protocol Across Interfaces: Web UI (recommended with dsh-web-ui) and dsh-tui (recommended with dsh-TUI) share database services and an eight-tool protocol; data mode retains sql-query, sql-write, sql-cmd, render-analysis, and Catalog-related tools without mixing in unrelated capabilities like describe_image or ssh_*.
  6. Support for Common Business Databases: MySQL, PostgreSQL, SQLite, Oracle, Hive, Impala, ClickHouse, Apache Doris, SQL Server, etc., covering business databases, analytical databases, local files, and data warehouse scenarios.
  7. Read-Only Protection and Password Handling: Supports read-only mode and database read-only accounts; temporary passwords are hidden in TUI and are not restored as form drafts.

Web UI also provides a “Database Workbench”: in the database button at the top right of the input box, you can complete connection configuration, browse databases and tables, data governance, and temporary SQL execution in the same pop-up.

Installation and Activation

The following commands are from the repository README. To install to the Web profile (npm method recommended):

dsh plugin --profile web add @yejiming/dsh-data-agent

Alternatively, install directly from GitHub:

dsh plugin --profile web add github:omdsh-dev/dsh-data-agent

If primarily using dsh-tui in the terminal, install to the corresponding profile:

dsh plugin --profile dsh-tui add @yejiming/dsh-data-agent

The plugin automatically installs the “Data Mode” preset and preloads relevant database tools when the profile starts. The lib/ directory is committed to the repository, so no local build is required when installing via npm or GitHub.

Typical Usage Examples

Using in Web UI

Start the Web UI:

dsh --profile web

Steps:

  1. Create a new session and select “Data Mode.”
  2. Click the database button at the top right of the input box and fill in connection information in the workbench.
  3. After a successful connection, directly ask analysis questions in the dialog box.
  4. Based on the first round of results, follow up to have DSH narrow the scope, compare dimensions, or summarize conclusions.

For example, you can enter:

Analyze order changes over the last 30 days, find the regions and products with the most significant sales declines, and explain the main reasons.

DSH will automatically check relevant tables, generate and execute queries, and complete the analysis based on actual results.

Using in dsh-tui

After starting the terminal interface, switch to data mode and connect to the database:

/preset data-agent
/database connect

After a successful connection, ask questions directly. Common database commands include:

/database status       Check current connection
/database test         Test current connection
/database disconnect   Disconnect current connection
/catalog scan          Start Catalog scan
/catalog status        View scan results

To obtain more valuable analysis, it’s recommended to add business objectives, time ranges, and focus dimensions to your questions, for example:

Analyze sales and gross margin changes by region in Q2 2026, identify regions with abnormal performance,
break down further by category and core customers, and provide three actionable business recommendations.

Prerequisites

When DSH executes queries, your local machine must be able to access the target database and have the corresponding command-line client installed (e.g., mysql for MySQL, psql for PostgreSQL, etc.). ClickHouse uses the plugin’s built-in HTTP adapter, so clickhouse-client is not required. It’s recommended to prepare a read-only database account to complete explorations without modifying business data.

Applicable Scenarios and Considerations

Who It’s For:

  • Data analysts or engineers who need to directly query databases, view data, and perform ad-hoc analysis within DSH conversations.
  • Business users who want to lower the SQL barrier using natural language while ensuring results are based on real query data.
  • Developers already using DSH Web UI or dsh-tui who want to add a “Data Mode” workflow to their agents.

Important Notes:

  • The plugin and its adapters run within the DSH process, accessing the database and file system with the current dsh process permissions, not at the OS-level sandbox. Before installation, you should review the source code and MIT license yourself to confirm compliance with your security policies.
  • It’s recommended to use a database read-only account and enable read-only mode in the connection form. When read-only is not enabled, the Agent can execute update or management statements as requested; before connecting to a production database, confirm account permissions and backup strategies.
  • Database connections in different sessions are isolated, making it convenient to handle different projects or client environments separately.
  • The Catalog only persists anonymized metadata and manually confirmed definitions, not passwords, query results, or sample rows; confirmation and deletion of AI candidates are done on the Web “Data Governance” page.
  • If you encounter failed to mount or cannot find @yejiming/dsh-data-agent, it’s usually because the current profile hasn’t installed the plugin or is still using an older preset. You need to execute the corresponding installation command, then completely exit and restart DSH.

Conclusion

If you want to expand DSH from a “code-writing agent” to a “data assistant that can connect to databases, query data, and generate reports,” dsh-data-agent is currently one of the most complete paths in the community: conversational questioning, automated SQL, data governance, HTML reports, and read-only protection are all integrated into the “Data Mode” preset, and Web and TUI can share the same capabilities.