Skip to main content

AI Agent (In-App Assistant)

Last Updated: March 27, 2026 (NQU-401 review) Last Reviewed: 2026-04-01

Overview​

The AI Agent is a context-aware assistant that helps you navigate Nquiry, understand features, and get guidance on investigation methodology. It's available throughout the application via a floating button.

Accessing the AI Agent​

Floating Button​

  • Located in bottom-right corner
  • Click to open the chat panel
  • Available on all pages
  • Shows context indicator

Chat Panel​

The panel slides in from the right:

  • Chat history maintained during session
  • Suggested prompts for quick access
  • Context-aware responses

What the AI Agent Knows​

Application Context​

The agent understands:

  • Current investigation details
  • Project status and progress
  • Available features and navigation
  • Your current location in the app

Investigation Context​

When inside a project, the agent knows:

  • Investigation title and focus
  • Number of topics, questions, evidence
  • Current phase (planning/collection/etc.)
  • Work type (investigation, audit, etc.)

Feature Knowledge​

The agent can explain:

  • How to use any feature
  • Best practices for investigations
  • Navigation and shortcuts
  • Troubleshooting common issues

What to Ask​

  • "Where do I upload evidence?"
  • "How do I generate a report?"
  • "Where can I see my analysis results?"
  • "How do I link evidence to questions?"

Feature Explanation​

  • "What are quality metrics?"
  • "How does semantic search work?"
  • "What's the difference between topics and questions?"
  • "How do I export my project?"

Methodology Guidance​

  • "What makes a good investigation question?"
  • "How should I organize my evidence?"
  • "When should I move to the analysis phase?"
  • "What should my report include?"

Current Project Help​

  • "What should I focus on next?"
  • "Which questions don't have evidence?"
  • "Am I ready to generate analysis?"
  • "What are my evidence gaps?"

Suggested Prompts​

The agent provides context-aware suggested prompts:

Based on Phase​

PhaseExample Suggestions
Planning"What makes a good question?"
Collection"What types of evidence should I collect?"
Analysis"How do I interpret the analysis results?"
Reporting"What sections should my report include?"

Based on Progress​

  • Few questions: "How do I create my first topic?"
  • No evidence: "How do I start collecting evidence?"
  • Evidence gaps: "Which questions have no linked evidence?"
  • Ready for analysis: "All evidence collected. Ready to analyze?"

Limitations​

What the Agent Cannot Do​

Cannot perform actions:

  • Upload or modify evidence
  • Generate analysis
  • Create investigations
  • Change settings

Cannot access:

  • Other users' data
  • Other organizations
  • System internals
  • External websites

Cannot provide:

  • Legal advice
  • Technical implementation details
  • Information about internal architecture
  • Specifics about AI models used

Privacy​

The agent:

  • Only sees your current investigation
  • Doesn't retain conversation history between sessions
  • Doesn't share information across organizations
  • Logs conversations for quality (anonymized)

Rate Limits​

Usage Limits​

  • 30 messages per minute
  • FREE from AI generation quota (NQU-417, deployed 2026-03-28)
  • 200/day soft cap per user
  • DB migration applied to remove quota counting

Quota Exemption​

The AI Agent is no longer subject to AI generation quotas. A database migration removed quota counting for chat messages, allowing unlimited AI guidance within the soft cap.

Best Practices​

Effective Questions​

Be specific:

  • "How do I link evidence to questions?" ✅
  • "Help with evidence" ❌

Provide context:

  • "I have 5 questions but no evidence yet. What should I do?" ✅
  • "What next?" ❌

Ask one thing at a time:

  • "What's the best way to organize topics?" ✅
  • "How do I organize topics, add questions, and upload evidence?" ❌

When to Use​

Good use cases:

  • Learning a new feature
  • Getting unstuck
  • Methodology questions
  • Quick navigation help

Better to use documentation:

  • Detailed technical information
  • Complete feature references
  • Compliance requirements

Technical Details​

Context Building​

The agent receives:

  • Investigation summary (title, status, counts)
  • Current page location
  • Topic and question lists (summarized)
  • Evidence overview
  • Guided context (gaps, progress)

Model​

  • Uses Claude Haiku 4.5 via AWS Bedrock (switched from Sonnet in NQU-357 for cost efficiency)
  • Separate from the analysis generation model (which uses Sonnet)
  • Optimized for fast, conversational assistance

Response Style​

The agent is designed to:

  • Be concise and actionable
  • Use markdown for clarity
  • Stay professional but approachable
  • Focus on helping you succeed

Troubleshooting​

Agent Not Responding​

  • Check internet connection
  • Refresh the page
  • Check AI quota (Settings → Billing)

Unhelpful Responses​

  • Try rephrasing your question
  • Be more specific
  • Provide more context about what you're trying to do

Context Seems Wrong​

  • Make sure you're in the right investigation
  • Refresh the page to update context
  • Check that data has been saved