Universal Deep Research: Bring Your Own Model and Strategy

Peter Belcak, Pavlo Molchanov
NVIDIA Research
Tech Report Code Lab
Universal Deep Research -- diagram of mechanism

An illustration of the UDR setup. The deep research agent takes both the research strategy and research prompt as inputs, compiles the strategy from natural language into an actionable research orchestration code snippet, and then carries out the strategy, delivering the final report to the user.

Abstract

Deep research tools are one of the most widely impactful and most numerous agentic systems of the present day. We observe, however, that each deep research agent introduced so far is hard-coded to carry out a particular research strategy using a fixed choice of tools.

We introduce Universal Deep Research (UDR), a generalist agentic system that wraps around any language model and enables the user to create, edit, and refine their own entirely custom deep research strategies without any need for additional training or finetuning.

To showcase the generality of our system, we equip UDR with example minimal, expansive, and intensive research strategies, and provide a user interface to facilitate experimentation with the system.

Uses & Deployment

UDR is presented as a prototype for a generalist agentic system that can be used to carry out deep research tasks in a variety of domains. The system is designed to be flexible and adaptable. Its most notable novel features include:

  • Custom Strategy Programming via Natural Language. UDR allows users to define and program their own research workflows in natural language, which the system then converts into executable, auditable code. This offers a practical route to productize user-authored agentic behaviors without needing new training or finetuning.
  • Model-Agnostic Research Tooling. UDR decouples research logic from the language model, enabling developers to wrap any LLM—regardless of vendor or architecture—into a functional deep research tool. This opens up product possibilities that mix-and-match best-in-class models with custom or domain-specific research strategies.
  • User-Controlled, Strategy-Driven Research Interfaces. The prototype demonstrates a UI that supports live strategy editing, selectable strategy libraries, progress notifications, and report viewing, offering a practical blueprint for building end-user products that combine transparency, modularity, and customizability.

BibTeX


@techreport{belcak2025universal,
  title={Universal Deep Research: Bring Your Own Model and Strategy},
  author={Belcak, Peter and Molchanov, Pavlo},
  year={2025},
  institution={NVIDIA Research},
  url={https://arxiv.org/abs/2509.00244},
}