What is node-DeepResearch
node-DeepResearch is an open-source AI agent project that continuously searches and reads web pages, using the Gemini language model and the Jina Reader tool to reason step by step and answer complex questions until it finds the answer or exceeds the token budget. The project supports multi-step reasoning and complex queries, and can handle tasks ranging from simple questions to complex multi-step reasoning. node-DeepResearch provides a Web Server API, making it convenient for users to submit queries via HTTP requests and get real-time progress.
Main features of node-DeepResearch
- Continuous search and reading: Searches for relevant information using search engines (such as Brave or DuckDuckGo) and reads web page content until it finds the answer to the question or exceeds the set token budget.
- Multi-step reasoning: Handles complex multi-step problems by breaking them down step by step and solving them gradually.
- Real-time progress feedback: Provides real-time progress updates based on the Web Server API, so users can keep track of the query's progress at any time.
- Flexible query methods: Supports everything from simple factual questions to complex open-ended questions, such as predicting future trends or formulating strategies.
Technical principles of node-DeepResearch
- Language model: Uses Gemini as the core language model, responsible for generating answers and reasoning logic. Gemini is a powerful general-purpose language model that can handle a variety of natural language tasks.
- Web page content processing: Uses Jina Reader to convert web page content into plain text format suitable for language model processing. Jina Reader is an open-source tool specifically designed for processing HTML web page content.
- Search and information extraction: Obtains relevant information using search engines (such as Brave or DuckDuckGo). The search engine provides web page links, and the agent extracts useful information by reading the web page content.
- Multi-step reasoning process：
- Initialization: Sets the initial context and variables.
- Budget check: Checks at each step whether the token budget has been exceeded.
- Generate prompt: Generates the input prompt for the language model based on the current question.
- Model generation: Calls Gemini to generate an answer or the next action.
- Action processing: Executes the corresponding operation based on the generated action (such as searching, visiting a web page, generating an answer, etc.).
- Result evaluation: Evaluates whether the generated answer meets the question's requirements; if not, continues reasoning.
- Loop and termination: If the answer is found within the budget, the query ends; if the budget is exceeded or no answer can be found, it enters "Beast Mode" to generate the final answer.
Project address of node-DeepResearch
- GitHub repository：https://github.com/jina-ai/node-DeepResearch
Application scenarios of node-DeepResearch
- Literature review: Helps researchers quickly find and organize literature in related fields and generate literature review reports.
- Market research: Quickly collects and analyzes market data, and understands competitor dynamics, industry trends, etc.
- News reporting: Quickly collect and organize background information and related reports on news events, helping journalists quickly write news articles.
- Learning Assistance: Help students answer complex academic questions and provide learning materials and reference answers.
- Policy Research: Help policymakers collect and analyze background information, implementation effects, etc. of relevant policies.