jina-ai/node-DeepResearch
Node DeepResearch
node-DeepResearch is an autonomous web research engine that uses large language models to iteratively search, read, and reason over web content to answer complex questions. It provides a chat-based interface that displays real-time reasoning steps and final answers, and can be configured to focus exclusively on academic papers by limiting searches to academic repositories.
The research engine operates through an agentic search-read-reason loop that repeatedly searches, reads, and reasons until a stopping condition is satisfied. It enforces a token budget to cap total consumption and failed attempts, guaranteeing a final answer. It ranks discovered URLs by combining relevance, authority, recency, and diversity, and selects the most relevant text segments from documents via chunked passage selection. For structured long-form reports, it iteratively builds an outline, searches per section, and refines for narrative coherence.
Beyond web text extraction, the engine extracts content from PDFs, generates textual captions for images on web pages using a vision language model, and formats search results for language model consumption. It can route reasoning steps to a local language model for structured output. The application is packaged as a Docker container with environment-variable configuration for portable deployment and also exposes an OpenAI-compatible API endpoint for programmatic client access.
Features
- Research Engines - Provides the core LLM-powered search-read-reason loop that defines the repository's purpose.
- Autonomous Web Research Loops - Operates an autonomous loop that searches, reads, and reasons over web content until a stopping condition is satisfied.
- Autonomous Web Researchers - Operates as an autonomous agent that iteratively searches and synthesizes web information.
- Reasoning Budget Controllers - Enforces token and failure budgets to cap consumption and guarantee a final answer before budget expiry.
- Web Research Agents - Implements an iterative search-read-reason loop with token budget control for multi-step queries.
- Search and Research - Queries search engines and returns top results with full content formatted for language model consumption.
- Chat Interfaces - Provides a chat interface that displays reasoning steps and final answers in real time.
- Academic Source Filters - Filters web searches to academic repositories, grounding answers on scholarly papers.
- Focused Retrieval - Provides a configurable mode that restricts research to academic paper sources.
- OpenAI-Compatible APIs - Exposes a standard chat completions endpoint compatible with the OpenAI API format for programmatic client access.
- Research Report Drafting - Constructs long-form reports by creating outlines, searching per section, and refining for narrative coherence.
- Local Model Integrations - Routes reasoning steps to a local language model and receives structured output for privacy or offline use.
- Local Model Structured Outputs - Routes reasoning steps to a local language model and receives structured output directly.
- Iterative Report Generators - Constructs structured reports by creating outlines, searching per section, and refining for narrative coherence.
- Passage Selection Strategies - Selects the most relevant text segments from a document using vector similarity comparison.
- Web Page Markdown Converters - Fetches web pages and returns their content as clean markdown text structured for language model ingestion.
- Web Content Fetching - Fetches web pages, extracts clean markdown text, and generates image captions for language model ingestion.
- Multi-Signal Ranking Strategies - Ranks discovered URLs by combining relevance, authority, recency, and diversity to prioritize reading.
- Web Chat Interfaces - Provides a web chat interface that displays reasoning steps and final answers in real time.
Star history
How this analysis was created: This summary and feature list were written by an AI model that read the project's README and public documentation pages. Each feature links to the documentation it came from; stars, license and language come straight from the GitHub API. The model does not read the source code, and the analysis is refreshed when the project is re-analysed. Learn more on our About page.
