u14app/deep-research
Deep Research
Deep research 是一个自动化研究生成系统，使用大型语言模型和网络搜索引擎来合成关于复杂主题的综合报告和深度分析。它将实时网络搜索结果与上传的本地文档相结合，使生成的内容基于特定的事实数据。
该系统采用迭代研究工作流，通过编辑、更新和重启特定研究阶段的分步过程来优化报告。它可以将非结构化的报告数据转换为知识图谱可视化，以映射不同发现之间的关系并组织研究项目的结构。
功能领域包括多个 AI 模型和搜索引擎的集成、使用专用模板定义研究范围，以及研究历史的管理。该系统还支持模型上下文协议 (Model Context Protocol)，以将研究能力和数据流连接到外部工具。
Features
- Structured Research Reports - Synthesizes comprehensive analysis reports by combining large language models with web search engines to aggregate information.
- Iterative Research Tools - Implements a multi-step web research system that generates queries and refines searches based on identified knowledge gaps.
- Web-Augmented Retrievers - Combines real-time web search results with uploaded local documents to ground model generations in specific facts.
- Local Document Indexing - Processes uploaded documents and files into searchable vector stores for RAG pipelines.
- AI Model Integrations - Connects multiple large language models and search engines to balance retrieval speed with analysis depth.
- Automated Research Generation - Uses large language models and web search engines to synthesize comprehensive reports and deep-dive analysis on complex topics.
- Research Report Drafting - Provides a process for iteratively retrieving and refining information to draft detailed research reports.
- Hybrid Knowledge Integration - Unifies user-provided documents and local resources as supplementary data sources for automated research.
- Language Model Connectivity - Integrates various language models and search engines to balance analysis depth with retrieval speed.
- Retrieval Iteration Loops - Repeatedly queries language models and search engines until a comprehensive information threshold is met.
- Document Search Grounding - Grounds AI generated reports in specific factual data from a combination of local documents and web search.
- Research Synthesis - Uses LLMs to synthesize raw scraped web data and uploaded local documents into structured research reports.
- Model Context Protocol Integrations - Uses a standardized communication protocol to connect core logic to external tools and disparate data sources.
- MCP Protocol Integrations - Uses the Model Context Protocol to connect research capabilities and data streams to external AI tools.
- Research Data Management - Stores previous results for review and triggers further in-depth study based on past reports.
- Process State Recovery - Tracks research progress in stages to allow restarting the synthesis loop from a specific saved checkpoint.
- Template-Driven Reporters - Guides the research process by applying pre-defined structural schemas for specific report types like market maps.
- Knowledge Graph Visualizations - Transforms unstructured report data into a structured network of nodes and edges to visualize conceptual relationships.
- Investigation Scoping - Sets the initial topic and applies specialized templates for market maps or technical deep dives to guide investigations.
