New to Deep Researcher Agent? This page walks you through the documentation in the order that will get you productive fastest.
Set up Python, install dependencies with uv, and configure your environment
variables (primarily NVIDIA_API_KEY).
Read: Installation
Launch the CLI and submit your first research query. This gives you a working mental model of what the system does before you look at how it works.
Read: Quick Start
Learn the two-path design — an intent classifier routes queries to either the fast shallow researcher or the multi-phase deep researcher — and how data flows through the system.
Read: Architecture Overview then Data Flow
Each agent has its own page covering state models, configuration, prompt templates, and internal flow diagrams.
- Intent Classifier — Query routing
- Shallow Researcher — Fast, bounded tool-calling
- Deep Researcher — Multi-phase subagent workflow
- Clarifier — Human-in-the-loop before deep research
Once you understand the agents, learn how to tailor the system to your needs:
- Swap LLMs — Use different models for different roles
- Enable or disable tools — Configure which data sources agents can access
- Edit prompts — Modify agent behavior through Jinja2 templates
- Add a new tool — Integrate a new search API or data source
- Configuration reference — Full YAML config guide
Move from local development to Docker Compose.
Read: Docker Compose then Production