Chat & Querying

The chat interface is where you interact with GenAI RAG. Type natural language questions, watch the agent reason and search in real time, and receive answers with inline citations linking back to your source data.

Introduction

GenAI RAG queries combine retrieval and generation: when you submit a question, an AI agent reasons about your intent, searches across the nexsets you have selected, and streams a natural language answer with inline citations. Each citation links back to the specific source data the agent used, so you can verify any claim in the response.

The chat preserves conversation history across turns within a session, so you can ask follow-up questions, refine results, and correct misunderstandings without restating context. For configuration of authentication, response tuning, and filter schemas, see Settings. For detailed source and tool-call exploration, see Canvas Panel.

The Chat Interface

The chat view consists of:

• Message list — scrollable area showing the conversation, auto-scrolls as new content arrives
• Chat input — auto-expanding text area at the bottom, with send and stop buttons
• Scroll-to-bottom button — click to jump back to the latest message
• Prompt chips — when the chat is empty, suggested questions appear based on the number of nexsets you have selected

Sending a Query

1. Type your question in the chat input.
2. Press Enter to send, or click the Send button.
3. Press Shift+Enter to insert a new line without sending.
4. If a requirement is missing (no credential or no nexset), a warning strip appears below the input telling you what to add before you can send.

Query Best Practices

Be specific — include date ranges, attributes, or parameters. Rather than "What are the sales figures?", ask "What are the Q3 2024 sales figures for the Northeast region?" Provide context and use follow-up queries to refine results.

A warning appears if your input is too long.

Streaming Responses

When you send a query, the agent processes it in real time. The Agent Timeline appears above the response, showing three phases:

1. Thinking (brain icon) — the model's chain-of-thought as it reasons about your question
2. Researching — parallel tool calls as the agent searches your nexsets
3. Generating — the model composes its final answer

The response streams in as it is generated.

Agent Timeline

The timeline visualizes what the agent is doing in real time:

Tool Call Cards

Each search operation appears as a card showing:

• Tool name and display name
• Status — Running (spinner), Complete, or Error
• Elapsed time
• Search query — the query the agent sent to the nexset
• Result count — e.g., "5 results"

Auto-Collapse

When the agent finishes researching and starts generating, the timeline auto-collapses into a compact summary — a single line like "Thought and searched Data Source A (5) and Data Source B (3) in 2.3s". Click it to expand the full timeline again, or to open the Run Summary in the canvas.

Thinking Display

On models that support extended thinking (e.g., Anthropic Claude), a collapsible Reasoning section appears showing the model's internal reasoning. This is separate from tool call activity and can interleave with it.

Citations

Answers contain inline citation markers rendered as superscript badges (e.g., ¹). These link to the specific source data used for that claim.

• Click a citation badge to open the Citation Detail view in the canvas panel
• Click the "N sources" pill below a response to open the Sources List view
• Citation badges appear at their first occurrence — subsequent references to the same source reuse the same badge

Message Actions

Hover over any finalized assistant message to reveal the action toolbar:

Action	Description
Copy	Copy the full message text to clipboard
Thumbs Up	Mark the response as helpful
Thumbs Down	Mark the response as not helpful
Regenerate	Re-send the preceding user message to get a new response

Multi-Turn Conversations

GenAI RAG maintains session context across turns. When you send a follow-up question, the agent has access to the prior conversation history, allowing you to:

• Refine previous answers ("Be more specific about Q2")
• Ask about related data ("What about the same metric for Q3?")
• Correct misunderstandings ("I meant the US region, not global")

Conversation history persists between sessions.

Starting Fresh

To begin a new conversation, click the Clear Chat button (trash icon) in the top bar and confirm, or press Ctrl/Cmd+Shift+N.

Connection Status

While streaming, the interface monitors the connection:

• Reconnecting... (yellow, pulsing) — brief interruption detected, automatically reconnecting
• Offline (red) — no events received for an extended period

If an error occurs during streaming, an error banner appears at the top of the chat with the error message and a Retry button to re-send the last query.

Stopping a Stream

While the agent is processing, the Send button becomes a red Stop button. Click it to abort the stream immediately. Any partial content already rendered remains visible but is not finalized.

Usage Information

Below each finalized assistant message, a footer shows:

• Output tokens — how many tokens the response consumed
• Tool calls — number of search operations performed
• Cost — estimated cost