michaelthwan/searchGPT

Grounded search engine (i.e. with source reference) based on LLM / ChatGPT / OpenAI API. It supports web search, file content search etc.

What it solves

searchGPT provides a way to prevent LLM hallucinations by grounding the model's answers in real-time factual information. It transforms a standard LLM into a search engine that provides natural language answers backed by specific sources, similar to the functionality of the new Bing.

How it works

The system retrieves information from two primary sources: the live Internet via the Azure Bing Search API and local files (such as PPT, DOC, and DOCX). It uses OpenAI embeddings for semantic search to find relevant content within those sources. This retrieved information is then fed into an LLM (such as OpenAI or GooseAI) to generate a grounded, factual answer based on the provided context.

Who it’s for

Developers and users who want to build or deploy a minimal implementation of a grounded search engine that combines web search and document search with LLM-generated summaries.

Highlights

  • Dual-Source Grounding: Combines real-time web search results and local file content search.
  • Semantic Search: Uses embeddings to retrieve the most relevant information from sources.
  • LLM Integration: Supports multiple providers including OpenAI and GooseAI.
  • Intuitive Interface: Includes a built-in frontend web application for easy interaction.

Related

  • Project
    Joe-B-Security/awesome-prompt-injection

    A curated, community-driven list of resources (articles, papers, tools, CTFs) about the prompt injection vulnerability in LLM applications, aimed at understanding, testing, and mitigating the risk.

  • Project
    swjybky/deepwrite

    A local-first AI writing workbench for creators that integrates specialized agents and a diff-based review system to collaboratively draft and refine manuscripts.

  • Project
    LifelongLazyLearner/qu-ai-wei

    A Simplified Chinese rewriting skill that removes stilted, formulaic AI-generated patterns to make text sound natural while preserving all original facts and technical terms.

  • Project
    datawhalechina/deep-learning-notes

    A self-published, continuously updated guide to deep learning with a private learner's narrative — moving from basic neural networks, PyTorch, CNNs, and RNNs, through Transformers and FlashAttention, into building GPT-2 from scratch, LLM training engineering, and modern inference frameworks (vLLM, SGLang).