Ask & Chat

The Ask page is the heart of ZenSearch - your primary interface for searching, chatting, and getting answers from your connected data sources.

Overview

The Ask interface combines powerful search with conversational AI, allowing you to:

• Ask questions in natural language
• Get AI-generated answers with citations
• Browse traditional search results
• Have multi-turn conversations
• Use AI agents for complex research

Interface Layout

Conversation Sidebar

The left sidebar manages your conversations:

• New Conversation: Start a fresh chat session
• Conversation List: Access previous conversations
• Timestamps: See when each conversation occurred
• Delete: Remove conversations you no longer need

The sidebar is collapsible and resizable to suit your preferences.

Main Content Area

The central area displays:

• Chat Mode: Messages with AI responses and citations
• Search Mode: Traditional search results with previews
• Agent Steps: Visual progress for agent tasks

Sources Panel

The right panel shows:

• Documents cited in the current response
• Relevance scores and coverage information
• Click any source to view the full document

Search Input

At the bottom:

• Multi-line text input for questions
• File upload button for attaching documents
• Send button with loading states
• Cancel button during streaming

Chat Mode

Starting a Conversation

1. Type your question in the input field
2. Press Enter or click Send
3. Watch as ZenSearch streams the response

Response Components

Each response includes:

• Answer Text: AI-generated content
• Inline Citations: Numbered references like [1], [2]
• Source Cards: Expandable document previews
• Confidence Indicator: A color-coded label (High / Medium / Low) showing answer trustworthiness based on source coverage, query classification, and quality checks
• Per-Claim Confidence: Individual claims within a response are annotated with their own confidence level (high, medium, or low) and linked to the specific sources that support them, so you can see exactly how well-grounded each statement is
• Metadata: Model used, token count

Streaming Source References

Sources appear as hoverable chips directly in the response as it streams, rather than only appearing after the full response completes. This gives you immediate visibility into which documents are being referenced without waiting for the entire answer.

• Source chips are inline and interactive during streaming
• Hover over a chip to preview the source document
• Click a chip to open the full source

Follow-up Suggestions

After each response, ZenSearch suggests relevant follow-up questions you can ask next. These appear as clickable chips below the answer, similar to the experience in tools like Perplexity.

• Suggestions are generated based on the current response and retrieved sources
• Click any suggestion to instantly send it as your next query
• Suggestions help you explore related topics and dig deeper into a subject

Canvas Artifacts Viewer

When an agent creates a canvas artifact (such as a report, analysis, or code snippet), it opens in a dedicated side panel alongside the conversation. You can view the content, browse version history, and see diffs between revisions without leaving the chat. See the Canvas & Artifacts page for full details.

Following Up

Continue conversations naturally:

You: What is our refund policy?
AI: Our refund policy allows returns within 30 days... [1]

You: What about digital products?
AI: Digital products have a modified policy... [2]

You: Can you summarize both in a table?
AI: Here's a comparison table...

Conversation Context

How Context Works

ZenSearch maintains context across your conversation to provide coherent, relevant responses. Each message builds on previous exchanges, allowing for natural follow-up questions.

Context Limits

Each AI model has a context window limit - the maximum amount of information it can consider at once. This includes:

• Your conversation history
• Retrieved source documents
• System instructions

Model	Context Window
GPT-4o	128,000 tokens
GPT-4o mini	128,000 tokens
Claude Sonnet 4	200,000 tokens
Claude Opus 4	200,000 tokens

Smart Context Management

ZenSearch automatically manages your context to ensure optimal responses:

• Source Priority: Retrieved documents take priority to ensure accurate answers
• Recent History: Most recent conversation turns are preserved
• Graceful Truncation: When context limits are reached, older messages are automatically summarized or removed

When Context is Full

For very long conversations, you may notice:

• Earlier messages are no longer referenced
• Responses focus on recent context
• A gentle reminder to start a new conversation

Tip: For fresh context on a new topic, start a new conversation using the New Conversation button.

Regenerating Responses

If an answer isn't helpful:

1. Click the Regenerate button
2. ZenSearch will search again and generate a new response
3. Compare different responses as needed

Search Mode

Switching to Search

Click the Search toggle to switch modes. Search mode provides:

• Traditional document results
• Relevance scores
• Preview snippets
• Faceted filtering

Faceted Search

Filter results using the sidebar:

Topics

Auto-extracted categories from your content:

• Technology
• Finance
• HR
• Sales
• etc.

Departments

Organizational groupings:

• Engineering
• Marketing
• Operations
• etc.

Languages

Filter by document language:

• English
• Spanish
• French
• etc.

Date Range

Filter by creation/modification date:

• Last 7 days
• Last 30 days
• Last 90 days
• Custom range

Sentiment

Content tone analysis:

• Positive
• Neutral
• Negative

Search Results

Each result shows:

• Title: Document name with link
• Preview: Relevant excerpt
• Source: Where the document came from
• Relevance: Score indicating match quality
• Tags: Topics and categories

Scope Selection

Choosing Collections

Use the scope selector to control search boundaries:

1. Click the Scope dropdown
2. Select specific collections or "All Collections"
3. See document counts per collection

Use Cases

Scope	When to Use
All Collections	General questions, company-wide search
Single Collection	Focused research in one area
Multiple Collections	Cross-functional queries

Agent Mode

Enabling Agents

Use the agent mode toggle:

• Auto: ZenSearch decides when to use agents
• Research: Always use agent with planning
• Off: Simple chat only

Agent Capabilities

Agents can:

• Search multiple collections
• Query connected databases
• Extract entities and relationships
• Perform calculations
• Synthesize information from many sources

Agent Progress Display

When an agent runs, you'll see:

[Planning] Breaking down your request...

[Iteration 1/5]
├── Thinking: Analyzing the question...
├── Tool: search_documents
│   Parameters: { query: "quarterly revenue" }
└── Result: Found 8 relevant documents

[Iteration 2/5]
├── Thinking: Need more specific data...
├── Tool: query_database
│   Parameters: { sql: "SELECT..." }
└── Result: 12 rows returned

[Synthesizing] Creating final response...

When to Use Agents

Agents excel at:

• Multi-part questions
• Comparative analysis
• Data aggregation
• Research tasks
• Complex reasoning

File Upload

Uploading Files

1. Click the Upload button (paperclip icon)
2. Select one or more files
3. Wait for processing to complete
4. Ask questions about the uploaded content

Supported Formats

• PDF documents
• Word documents (.docx)
• Text files
• Images (for visual questions)
• Spreadsheets

Upload Status

Files show status indicators:

• Uploading: Transfer in progress
• Processing: Extracting content
• Ready: File is searchable
• Failed: Upload error

Suggested Questions

When you first visit or have no messages, ZenSearch suggests questions:

• Based on your connected data sources
• Covering common use cases
• Click to instantly search

If you have no data sources connected, suggestions guide you to add connectors.

Keyboard Shortcuts

Shortcut	Action
Enter	Send message
Shift + Enter	New line in input
Cmd/Ctrl + K	Focus search
Escape	Cancel current request

Best Practices

Writing Good Questions

1. Be specific: Include context and details
2. Use natural language: Ask complete questions
3. Mention timeframes: "in the last quarter", "for 2024"
4. Reference specifics: Department, project, or document names

Examples

Poor Question	Better Question
"sales"	"What were our total sales in Q4 2024?"
"policy"	"What is our work from home policy?"
"how to"	"How do I submit an expense report?"

Understanding Citations

• [1], [2], etc.: Click to view source document
• Multiple citations: Information synthesized from several sources
• No citations: General knowledge or reasoning

Troubleshooting

No Results Found

1. Check that collections contain relevant content
2. Try broader search terms
3. Verify the correct scope is selected
4. Wait for syncs to complete

Slow Responses

1. Complex questions take longer
2. Agent tasks require multiple steps
3. Large document sets need more processing

Conversation Too Long

If you notice the AI isn't referencing earlier parts of your conversation:

1. This is normal for very long conversations
2. The AI prioritizes source documents and recent messages
3. Start a new conversation for fresh context
4. Consider breaking complex topics into separate conversations

Incorrect Answers

1. Check the cited sources for accuracy
2. Try rephrasing your question
3. Use agent mode for complex topics
4. Verify source documents are up to date

Next Steps

• Agents - Create custom AI agents
• Canvas & Artifacts - Persistent content objects
• Knowledge - Manage data sources
• Search Deep Dive - Advanced search techniques