6. AI Chat & Analysis

Getting Started with AI Chat

The AI chat interface lets you analyze your data using natural language questions. After importing data, ask questions as if you were talking to a data analyst. RangerIO answers locally by default; when you opt in to the Stage 2 connector, it routes the dossier (not the raw rows) to a cloud model and masks any remaining sensitive values before they leave your workstation.

💡 UI Location: Find the chat interface by clicking the "Chat" or "AI Assistant" tab in the left sidebar after importing data

How AI Chat Works

Processing Pipeline:
1. Question Analysis - Intent detection and entity extraction
2. Context Retrieval - Hybrid search (vector + BM25) over the dossier and indexed chunks
3. Data Query - SQL execution against the local analytics engine if needed
4. Prompt Engineering - Combine context with user question
5. Safe Handoff (when routing to cloud) - Mask sensitive values before any cloud round-trip; restore them when the answer returns
6. Inference - Generate the answer locally, or via the Stage 2 connector if enabled
7. Citation Linking - Add source references

Asking Questions

Basic Queries

• "How many rows are in this dataset?"
• "What's the average sales amount?"
• "Show me the top 10 customers by revenue"
• "List all unique product categories"

Complex Analysis

• "What trends do you see in sales over time?"
• "Are there correlations between price and quantity?"
• "Identify outliers in revenue"
• "Summarize key patterns in customer behavior"

Data Transformation

• "Convert dates to YYYY-MM-DD format"
• "Remove rows where revenue is negative"
• "Create a profit margin column"
• "Find and remove duplicates"

RAG Modes

Basic Mode - Fast, simple queries (2-5 seconds)
- Quick lookups and filtering
- Best for straightforward questions

Assistant Mode ⭐ Recommended (5-15 seconds)
- General analysis and multi-step reasoning
- Balanced speed and accuracy
- Best for most use cases

Deep Search Mode - Comprehensive analysis (15-60 seconds)
- Complex multi-source queries
- Maximum context and accuracy
- Best for research-style questions

Understanding Responses

AI answers include:
- Direct answer to your question
- Supporting evidence and statistics
- Source citations with links
- Confidence indicator (🟢 High, 🟡 Medium, 🔴 Low)

Best Practices

DO:
- ✅ Be specific: "Show Q1 2024 sales" not "Show sales"
- ✅ Use context: "Based on customer table..."
- ✅ Ask follow-ups to build on previous answers
- ✅ Specify output format: "as a table"

DON'T:
- ❌ Be vague: "Tell me about the data"
- ❌ Ask multiple unrelated questions at once
- ❌ Expect predictions beyond your data

Working with Multiple Sources

• Select multiple data sources from dropdown
• AI searches across all selected sources
• Citations show which source provided each insight
• Use "All Sources" for comprehensive searches

Performance Tips

For faster responses:
- Use Basic mode
- Select specific data sources (not "All")
- Use smaller models
- Ask focused questions

For better accuracy:
- Use Deep Search mode
- Step up to a larger local model (e.g., Qwen3 8B / DeepSeek-R1 Distill 7B), or route via the Stage 2 connector to a cloud model
- Provide more context
- Request confidence levels

Chat History

• All conversations saved per project
• Browse by date or search
• Click to restore conversation
• Export chat to Markdown/PDF/HTML

For more details, see the in-app help and FAQ section.