Deep Writing

​

Feature overview

Deep Writing is a deep research feature built on an open-source multi-agent framework. This feature deeply integrates with your business data, focusing on data analysis, toolchain construction, and output format design to deliver an end-to-end intelligent workflow for generating research reports. It can automatically create various content products, such as daily monitoring reports for new media accounts, industry market analysis reports, and documentary scripts.

​

Feature access

Navigate to Miaobi. The feature is available under Content Creation Agent > Deep Writing. On the Deep Writing page, the main heading reads, "Multi-agent assistance for producing complex long-form content." Below this is a task input box where you can configure parameters such as Writing Source, sub-agent, and methodology. The top of the page contains links to History and sub-agent information and management.

​

Feature details

​

Configuration page

Configure the settings, enter a writing topic, and then click to start the process. The following sections describe the configuration options:

• Upload local materials
  • Upload up to 10 reference documents. Supported formats are TXT, DOC, and PDF, with a maximum size of 10 MB per file.
  • Upload up to 3 data tables. Supported formats are CSV and XLSX, with a maximum size of 10 MB per file.
• Writing Source
  • By default, Internet Search is available. You can add specific URLs when searching the internet.
    In Internet Search, you can add specific websites and enable the Limit to Authoritative Data Sources toggle. When enabled, Deep Writing retrieves information only from your trusted sites, ensuring that content sources are authoritative, reliable, and controllable.In the left-side navigation pane, click Configuration and select the Writing Source tab. In the Internet Search data source, open the Specify Websites dialog box, enter or paste the target URLs (for example, aliyun.com or baidu.com), and click Confirm Add.
  • You can add a specific data source by either uploading local documents or importing data through an API.
  • Authorize an OSS data source.
    For instructions on how to add a specific data source or authorize an OSS data source, see the Create a data source document.
• Sub-agents
  • Project Manager: Plans tasks, analyzes the user's task goals, breaks down goals into sub-tasks for other agents, and reviews the results at each stage.
  • Data Collector: Searches the internet, reads materials, and organizes useful information into documents that serve as source material for subsequent analysis.
  • Report Writer: Summarizes existing materials to write a clear report for the user.
  • Data Analyst: Analyzes both offline and online search data, and creates charts based on the analysis results.
    When the Data Analyst sub-agent is enabled, the system provides two chart presentation options: you can choose to generate charts only, or to output both charts and the original data tables.Click the gear icon in the upper-right corner of the page and select the desired presentation option from the drop-down menu.
  The Project Manager, Data Collector, and Report Writer are mandatory sub-agents and are always enabled. The Data Analyst is an optional sub-agent that you can enable or disable with a toggle.
• Methodology: The system provides official methodologies for desk research reports, market analysis reports, consumer insight reports, and documentary scripts. You can also customize methodologies to create other types of reports.
• Table of contents: When this toggle is enabled, a table of contents is automatically generated at the beginning of the report. You can click a chapter title to jump to the corresponding content. If you enter an instruction in the input area (for example, "The generated article must have a table of contents"), the system automatically recognizes your intent and displays the Table of contents toggle below the input box. Enable this toggle to include a table of contents in the generated article. Click the send button on the right to submit the task.

​

Results page

1. The following sections describe the modules on the results page.

• The left panel shows the task information. The top shows your research request. The middle section contains the Execution Plan module, which lists all data collection and analysis steps, marking completed steps with a green checkmark. The bottom section displays the configuration for the current task (number of data sources and sub-agents) and provides download links for the generated output files, such as final_report.md.
• The right panel shows the results across four tabs:
  • Report: This tab displays the generated report and contains a verification view and a publication view. The verification view allows you to inspect the report's data and its sources online, retaining all intermediate files for traceability. The publication view shows the final, downloadable version of the report, with intermediate files removed. You can view a summary of the report, including sections like analysis purpose and scope, prediction logic, and their referenced data source tags.
  • Activities: This tab displays the collaborative workflow of the agents (such as Project Manager and Data Collector) during task execution. It includes steps like task allocation, reasoning and analysis, internet searches, and writing intermediate files, allowing you to trace the entire report generation process.
  • Files: This tab lists the files generated and used during the task. You can preview and download individual files. In the verification view, clicking an intermediate file (such as market_forecast_industry_growth_saas.md) opens a dialog box to view its content. The original search results area at the bottom of the dialog box shows source information, including the URL, publication time, site name, and title, which helps you trace data sources.
  • Data source: This tab lists all web page sources that were searched and read for the task. It includes source type identifiers and the corresponding URLs, making it easy to trace the original sources of the report's data.

1. Traceability and verification: In the verification view, all generated content is linked to traceable intermediate files. You can examine these files to see their original sources, allowing you to fully trace and verify the content's authenticity. Intermediate file names (for example, market_scope.md) are cited inline within the report body. Click a file name to view its content. The right sidebar displays a numbered list of original data sources, allowing for item-by-item traceability.
2. AI Reading Guide: For each report, the system automatically generates a content overview, key takeaways, keywords, a quick-read summary, and a mind map. It also supports a Q&A feature based on the report content, helping you read faster, understand deeper, and ask more precise questions. On the results page, the left panel contains a button to open the AI Reading Guide. The guide uses a two-column layout: the report content is on the left, and a panel with summaries and interactive features is on the right.
3. Download report: You can export the report in Markdown, PDF, or ZIP archive (including all resources) formats.