Developer Guide#

New to AI-Q? This page walks you through the documentation in the order that will get you productive fastest.

1. Install#

Set up Python, install dependencies with uv, and configure your environment variables (primarily NVIDIA_API_KEY).

2. Run the Agent#

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

3. Understand the Architecture#

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

4. Explore Individual Agents#

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

5. Customize and Extend#

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

6. Deploy#

Move from local development to Docker Compose.

Read: Docker Compose then Production