Open Deep Search: Unlocking Advanced AI Search for Everyone

Imagine if the power behind today’s smartest search engines was no longer limited to tech giants. That’s the promise of Open Deep Search (ODS), a breakthrough open-source framework that puts advanced AI-powered search and reasoning tools into the hands of everyone.

What Sets ODS Apart?

Unlike proprietary systems that keep their algorithms and data locked away, ODS is designed for openness and adaptability. It can connect with any large language model (LLM), whether you prefer open options or commercial APIs, giving you unmatched flexibility.

• Versatile integration: Use your choice of LLM—such as Llama3.1-70B or DeepSeek-R1—and upgrade easily as new models are released.
  
  
• Core components:
  • Open Search Tool: Retrieves diverse, high-quality information using smart techniques like query rephrasing and metadata formatting to maximize relevance.
  • Open Reasoning Agent: Interprets queries, orchestrates tool usage, and enables multi-step reasoning with frameworks like ReAct and CodeAct.
    
    
• Advanced agentic behavior: ODS agents decide when and how to use web search, calculators, or code interpreters, tailoring their strategy to each query’s complexity.

Solving the Problems of Closed AI Search

Closed-source search AIs limit transparency, slow innovation, and make it tough to audit algorithms for fairness or bias. ODS addresses these challenges head-on by providing a fully open, inspectable, and collaborative framework. Developers and researchers can scrutinize every step, suggest improvements, and drive the technology forward together.

Its agent-driven, step-wise approach ensures factual accuracy and reliable reasoning, rather than leaving these outcomes up to chance or hidden algorithms.

Performance That Rivals Tech Giants

• State-of-the-art accuracy: ODS-v2, powered by the CodeAct agent and DeepSeek-R1, leads on the FRAMES benchmark with 75.3% accuracy—beating top closed platforms like GPT-4o Search Preview.
• Strong in simple QA tasks: Achieves 88.3% accuracy on the SimpleQA benchmark, narrowing the gap with industry leaders and outpacing Perplexity’s solutions.
• Intelligent resource use: ODS agents increase the number of searches for complex, multi-hop questions, showcasing effective problem decomposition.
• Component value: Ablation studies confirm that each core module—especially the Open Search Tool and agentic control—drives significant performance gains.

Technical Advances Driving ODS

• Refined search integration: ODS reformulates queries, organizes results, and enhances context far beyond what simple open-source tools can do.
• Agentic reasoning frameworks: ReAct-based ODS-v1 interleaves reasoning and action; CodeAct-based ODS-v2 empowers the LLM to generate and execute code for complex problem solving.
• Few-shot prompting: By providing in-context examples, ODS agents boost accuracy on challenging, multi-step queries.
• Transparent modularity: Every component is open for audit, improvement, and integration with new models, ensuring ODS evolves as the ecosystem grows.

Shaping the Future of Search AI

With its open-source release, ODS invites the community to innovate, scrutinize, and collaborate. By breaking the lock of proprietary systems, it accelerates progress and collective understanding, vital for addressing bias, hallucination, and misuse in generative AI.

ODS’s plug-and-play design ensures it remains on the cutting edge as new language models emerge. The result? A democratic, trustworthy, and high-performance framework for advanced AI search that empowers everyone, not just tech giants.