This project provides an AI-powered conversational search engine that supports deep research capabilities, similar to Perplexity AI. It targets users who want to build a custom, LLM-driven search experience, integrating with various LLM providers, local models via Ollama/LMStudio, and the SearXNG meta-search engine. The primary benefit is a flexible, extensible search platform with advanced research features.

How It Works

The system orchestrates a multi-step "Deep Research" workflow. It analyzes user queries, generates clarifying follow-up questions, executes search queries across integrated engines, scrapes web content (using Jina.ai), and recursively refines the search based on results. Finally, it synthesizes this information into a comprehensive report using LLMs. This iterative, LLM-guided approach allows for deeper exploration and more detailed answers than traditional search.

Quick Start & Requirements

• Install/Run: Docker is recommended.
  • docker pull docker.cnb.cool/aigc/aisearch
  • Clone repo: git clone https://github.com/yokingma/search_with_ai.git
  • Configure .env.docker with API keys (e.g., OPENAI_KEY).
  • Run with docker compose up -d.
• Prerequisites: Docker, Node.js >= 20 (for development). API keys for LLM providers and optional search services.
• Resources: Requires API keys and potentially significant LLM token usage for deep research.
• Links: Live Demo, English README, 中文 README, 日本語 README

Highlighted Details

• Supports "Deep Research" workflow for comprehensive report generation.
• Integrates with numerous LLM providers (OpenAI, DeepSeek R1, etc.) and local models (Ollama, LMStudio).
• Built-in support for SearXNG (no key required), Bing, Google, and Tavily search engines.
• Customizable UI with light/dark mode and mobile support.
• Supports continue Q&A, caching, image search, and internationalization.

Maintenance & Community

The project is actively maintained by yokingma. Community links are not explicitly provided in the README, but the repository is hosted on GitHub.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive MIT license allows for commercial use and integration with closed-source applications.

Limitations & Caveats

The "Deep Research" feature is noted to be token-intensive and requires function-calling support. Web content extraction via Jina.ai is limited to 20 RPM without a JINA KEY.