AI optimization determines what content your Knowledge Base can understand, search, and retrieve effectively. This configuration controls how data is processed, which fields are indexed, and how system performance is balanced.
Use this guide to:
Select the right fields for indexing
Choose the appropriate processing mode
Understand how these decisions affect speed, cost, and retrieval quality
Field selection and processing configuration directly impact:
Search Relevance — AI retrieves better results when focused on meaningful fields
System Efficiency — Unnecessary fields increase processing time and cost
Content Explainability — Optimized fields make AI responses easier to trace and validate
Avoid treating every field equally. The goal is to prioritize semantic content and skip system noise.
Click the lightbulb icon in the right column of any Knowledge Table to open the Knowledge Configuration panel.
1. Review Detected Fields
All fields from your data source are listed with sample values for context.
2. Select Fields for Optimization
Check the fields that should be indexed and vectorized. These will be included in the AI search layer.
3. Choose Processing Mode
Select either:
Standard: Immediate processing with progress tracking
24-Hour: Background processing for large datasets (over 10k records)
4. Save or Reset
Use Save to apply the configuration or Reset to clear and revert to defaults.
Focus on fields that provide semantic context or describe content users will search.
Examples:
Long-form text: description, body, comments
Labels or categories: tags, themes
Titles or names: product_name, report_title
Avoid fields that don’t add meaning or introduce noise.
Examples:
System metadata: status, created_by
Numeric-only values: price, quantity
Timestamps: created_at, updated_at
Identifiers: ID, UUID, record_ref
Start with 3–5 core fields
Use the sample data preview to validate selection
Prioritize longer text fields for richer embeddings
Monitor completion time before scaling up
Revisit after observing search behavior or user feedback
Default mode for small to medium datasets. Runs in real-time with live progress display.
Best for:
Collections under 10,000 records
Iterative workflows or frequent updates
Immediate QA and tuning
Behavior:
Immediate start
Faster processing
Can be monitored from the interface
Background mode for large or complex datasets. Prioritizes scale over interactivity.
Best for:
Datasets exceeding 10,000 records
Collections with many long-form fields
Batch uploads or backfills
Behavior:
Processes in the background
Completion time up to 24 hours
Email notification sent when complete
Each available field includes:
A checkbox to enable/disable processing
A sample value to show real content
A real-time counter for selected fields
Use this preview to spot:
Irrelevant fields (e.g., “$0”, “2022-01-01”)
Duplicate or alias fields
Redundant system data
Use the blue Select All checkbox to bulk include or exclude all fields.
Enable or disable 24-Hour Processing Mode using the toggle at the bottom of the panel.
Save — Apply field selection and processing mode
Reset — Clear all selections and revert to default state
After configuration is saved, the system begins processing immediately (or in the background for 24-hour mode). Status is shown in the system panel or via email.
Example:
2,847 Total Chunks — Number of content segments parsed
2,847 Vectorized — Number of AI-processed segments
4 Optimized Fields — Fields selected for AI indexing
Use this mental model:
"If a user asks a question, can this field help answer it?"
If not, skip it.
Example:
A field like Approval Notes is useful for understanding why something was approved.
A field like Approval Date is not helpful for semantic retrieval.
Think of your Knowledge Base like a custom internal search engine. The goal is to include content-rich, explainable data—and ignore system noise.