OpenDeepSearch is an open-source search tool designed to enhance AI agent capabilities by providing deep web search and retrieval. It aims to democratize search by leveraging open-source reasoning models and agents, offering superior performance on complex, multi-hop queries compared to closed-source alternatives.

How It Works

OpenDeepSearch employs a dual-mode approach: Default Mode for quick, SERP-based results and Pro Mode for in-depth, semantically reranked results via web scraping. It integrates with AI agent frameworks like SmolAgents and utilizes LiteLLM for flexible LLM provider integration. Semantic search is powered by rerankers such as Qwen2-7B-instruct or Jina AI, with options for Serper.dev or self-hosted SearXNG for initial search.

Quick Start & Requirements

• Installation: pip install -e . or uv pip install -e .
• Prerequisites: PyTorch (torch) is required. API keys for search providers (Serper.dev or SearXNG) and rerankers (Jina AI) are necessary. LiteLLM requires API keys for chosen LLM providers (OpenAI, Anthropic, Google, etc.).
• Setup: Configure API keys via environment variables (e.g., SERPER_API_KEY, JINA_API_KEY, OPENROUTER_API_KEY).
• Demo: Run python gradio_demo.py for an interactive interface.
• Docs: https://arxiv.org/abs/2503.20201 (Paper)

Highlighted Details

• Outperforms closed-source alternatives on multi-hop queries (FRAMES benchmark).
• Supports semantic search reranking with models like Qwen2-7B-instruct and Jina AI.
• Integrates with SmolAgents and LiteLLM for seamless AI agent workflows.
• Offers two modes: Default (fast, SERP-based) and Pro (deep, semantically reranked).

Maintenance & Community

The project is associated with multiple authors from academic institutions, suggesting a research-driven development. Contact is available via GitHub issues.

Licensing & Compatibility

The project is released under an unspecified license. The citation lists multiple authors and institutions, and the arXiv link points to a paper, indicating a research project. Further clarification on licensing is needed for commercial use.

Limitations & Caveats

The README does not explicitly state the license, which is crucial for determining commercial usability. While performance claims are made, specific benchmarks beyond single-hop and multi-hop query comparisons are not detailed.