MindSearch is an open-source, LLM-based multi-agent framework for building a web search engine in the style of Perplexity.ai Pro and SearchGPT. Rather than stuffing one prompt with search results, it plans a question the way a human researcher would โ breaking it into sub-questions, exploring them in parallel, and synthesizing an answer. The project comes from the InternLM team, has earned about 6,900 stars on GitHub, is Apache 2.0 licensed, and was actively developed into 2025.
MindSearch models web research as a graph. A planner agent decomposes the user’s query into atomic sub-questions and arranges them as nodes in a WebSearchGraph; searcher agents then work through the graph, querying the web, reading pages and summarizing findings for each node. Because independent sub-questions run in parallel, the system can gather far more information in the same wall-clock time than a single sequential search agent, and the graph structure keeps multi-hop reasoning organized. The framework plugs into InternLM and other LLMs, uses search APIs such as Bing or DuckDuckGo under the hood, and ships with a ready-made web UI so you can run your own Perplexity-style answer engine.
Free. MindSearch is open source under the Apache 2.0 license; you only cover the cost of any LLM or search API keys you connect to it.
Developers and researchers building AI search products, research assistants, or answer engines who want an open, inspectable multi-agent architecture instead of a closed commercial API.