🚀 Try Zilliz Cloud, the fully managed Milvus, for free—experience 10x faster performance! Try Now>>

Milvus
Zilliz
  • Home
  • AI Reference
  • What input formats can DeepResearch accept beyond a simple text query (for example, an outline or a partial draft)?

What input formats can DeepResearch accept beyond a simple text query (for example, an outline or a partial draft)?

DeepResearch accepts multiple structured input formats beyond plain text queries, designed to help developers integrate the tool into various workflows. The system can process outlines in Markdown or bullet-point formats, partial drafts with placeholders or annotations, and structured data like JSON or YAML. For example, a user could submit a Markdown outline with headers like ## Introduction, ### Key Findings, and ### Next Steps, and DeepResearch would generate detailed content for each section. Similarly, JSON inputs could define parameters such as {"topic": "machine learning", "subtopics": ["neural networks", "training data"]}, allowing the tool to organize research around specific themes. These formats enable users to define scope and hierarchy upfront, guiding the tool’s output.

Developers can also submit partial drafts containing incomplete sections, code snippets, or inline comments. For instance, a draft might include a Python function with a # TODO: optimize this loop comment, prompting DeepResearch to suggest algorithmic improvements or alternative libraries. Similarly, a technical document with placeholder text like [Insert API example here] could trigger the tool to generate code samples or API documentation. This is useful for iterative workflows where users want to refine existing content rather than start from scratch. Partial drafts often include mixed formats—such as text interspersed with tables, diagrams in SVG markup, or mathematical equations in LaTeX—which DeepResearch parses to maintain context and coherence.

For technical users, DeepResearch supports domain-specific formats like CSV/TSV data tables, API request templates, or version control diffs. A CSV file containing experimental results could be analyzed to generate statistical summaries or visualizations. API templates (e.g., a cURL command with missing headers) might prompt the tool to suggest authentication methods or error-handling logic. Even Git diffs are usable inputs: submitting a code change snippet could trigger a review for potential bugs or performance issues. These formats allow developers to integrate DeepResearch into CI/CD pipelines, data analysis scripts, or documentation generators, making it adaptable to specialized use cases without requiring manual text conversion.

Like the article? Spread the word