langchain-ai/local-deep-researcher

What it solves

It provides a way to perform deep, iterative web research using entirely local LLMs, eliminating the need for cloud-based AI services for the reasoning and synthesis parts of the research process.

How it works

The assistant follows a cyclical research loop:

1. Query Generation: It uses a local LLM (via Ollama or LMStudio) to create a web search query based on a topic.
2. Information Gathering: It searches the web using tools like DuckDuckGo, Tavily, SearXNG, or Perplexity.
3. Summarization: The LLM summarizes the search results.
4. Reflection: The LLM analyzes the summary to identify knowledge gaps.
5. Iteration: It generates new queries to fill those gaps and repeats the process for a user-defined number of cycles.
6. Final Output: It produces a final markdown summary with citations for all sources used.

Who it’s for

Users who want a research assistant that runs locally for privacy or cost reasons, and developers looking to implement iterative research agents using LangGraph.

Highlights

• Local LLM Support: Compatible with Ollama and LMStudio.
• Flexible Search: Supports multiple search providers including DuckDuckGo (no API key required), Tavily, SearXNG, and Perplexity.
• Iterative Reasoning: Uses a reflection-based loop to identify and fill knowledge gaps rather than performing a single search.
• Visualized Workflow: Integrates with LangGraph Studio to allow users to visualize the research process in real-time.