DeepResearch supports multiple input formats designed to help users refine and expand their work beyond basic text queries. These include structured outlines, partial drafts, markdown-formatted content, code snippets, tabular data (CSV/TSV), and URLs pointing to external content. Each format is processed to extract context, identify gaps, and generate targeted outputs based on the provided structure or data.
For example, an outline with bullet points or section headings allows DeepResearch to analyze the hierarchy and suggest detailed content for each segment. A partial draft might include incomplete paragraphs or placeholder comments (e.g., “Expand this section on API design”), which the system uses to prioritize research or fill in missing information. Markdown formatting (headers, lists, code blocks) helps the tool distinguish between technical details, narrative text, and data examples. Code snippets or configuration files (e.g., YAML, JSON) can be parsed to identify dependencies or propose optimizations. Structured data in CSV/TSV formats enables statistical analysis or visualization recommendations. URLs to articles, documentation, or repositories allow DeepResearch to incorporate external sources into its output.
The system processes these inputs by first identifying their structure (e.g., hierarchical outlines, tabular data) and then applying context-aware algorithms. For instance, a draft with TODOs triggers task-specific research, while a CSV might prompt data summarization. Developers can combine formats, like embedding code snippets within a markdown draft, to create hybrid inputs that guide output precision. This flexibility ensures users can start from varying stages of their workflow without reformatting content.