Assistants

AI-powered chat assistants that answer questions using your knowledge base

What are Assistants?

Assistants are AI chat agents that answer questions based on documents in your Libraries. Each assistant is configured with specific behavior instructions and connected to relevant Libraries, allowing users to have natural language conversations about your data.

Think of assistants as specialized team members—each one knows about specific topics and can help users find information, answer questions, and provide insights based on your organization's documents.

Knowledge-Grounded

Answers based on your actual documents, not generic AI knowledge

Customizable Behavior

Configure how the assistant responds with instruction base cases

Multi-Library Access

Connect to multiple Libraries for comprehensive knowledge

Creating an Assistant

Basic Info

Name, description, icon
Select AI Model

Choose language model
Link Libraries

Connect knowledge sources
Configure Behavior

Add instruction base cases
Add Participants

Grant user access

Basic Configuration

Name	Required A descriptive name for your assistant (e.g., "Product Documentation Expert", "HR Policy Helper", "Technical Support Bot")
Description	Optional Brief description of what this assistant helps with. Visible to users when selecting an assistant.
Icon	Optional Upload a custom icon image (PNG, JPG) to visually identify this assistant
Language Model	Optional • Defaults to system default Select which AI model powers this assistant. Available models depend on your AI Services configuration. Different models have different capabilities—larger models are more capable but slower.

Linking Libraries

Assistants need access to Libraries to answer questions. Each linked Library becomes part of the assistant's knowledge base.

How Library Linking Works

Select which Libraries this assistant can search
For each Library, describe "Used For" — what topics/information this Library provides
When a user asks a question, the assistant searches linked Libraries for relevant content
The assistant uses found documents as context to generate accurate answers

Assistant: "Technical Documentation Helper"

"Engineering Docs"

"API references, architecture diagrams"

"Release Notes"

"Version history, changelog, upgrade guides"

Library Access Required

Users must have read access to the linked Libraries to use the assistant. If a user doesn't have access to a Library, that Library's content won't be included in search results for them.

Behavior Configuration (Base Cases)

Base cases are instruction rules that control how the assistant behaves and responds to different types of questions.

What are Base Cases?

Each base case is a conditional instruction that tells the assistant how to respond when certain conditions are met.

Field	Description
Condition	When to apply this instruction (e.g., "user asks about pricing", "question is in German")
Instruction	What the assistant should do (e.g., "Provide pricing details and include a link to the pricing page")
Sequence	Order of priority (lower numbers = higher priority). First matching base case is applied.

Example Base Cases:

Condition: "User asks about API authentication"

Instruction: "Explain OAuth 2.0 flow and provide code examples from the API documentation. Always include links to relevant sections."

Condition: "User needs troubleshooting help"

Instruction: "Ask clarifying questions about their environment (OS, version, error message) before providing solutions. Reference troubleshooting guides."

Condition: "Question is outside documentation scope"

Instruction: "Politely explain that you can only answer questions about documented topics. Suggest contacting support for other inquiries."

All Fields Optional

Base cases are optional. Without base cases, the assistant uses default behavior: answer questions directly using content from linked Libraries.

Participants & Access Control

Control who can use this assistant by managing participants:

Access Roles

Owner:

• Full control over the assistant
• Can edit all settings
• Can manage participants
• Can link/unlink Libraries
• Can delete the assistant

Participants:

• Can start conversations with the assistant
• Can ask questions and receive answers
• Cannot modify assistant configuration
• Cannot see other users' conversations

Add Participants

Navigate to Assistant Settings → Participants → Add Users to grant access to additional team members.

How Assistants Work

When a user asks a question in a conversation:

1. Semantic Search

The assistant searches all linked Libraries using vector similarity search to find relevant document chunks

2. Context Assembly

Retrieved document chunks are assembled as context, along with conversation history and any applicable base case instructions

3. Answer Generation

The AI model generates an answer based on the retrieved documents and instructions, citing sources from your Libraries

4. Response with Sources

The answer is shown to the user along with references to source documents, allowing verification and further reading

Common Use Cases

Customer Support Bot

Answers customer questions based on product documentation, FAQs, and troubleshooting guides

Libraries: Product Docs, FAQs, Troubleshooting

Base Case: "Always be polite and ask follow-up questions to clarify issues"

Internal HR Assistant

Helps employees find HR policies, benefits information, and answers to common HR questions

Libraries: HR Policies, Benefits Handbook, Onboarding

Base Case: "For sensitive topics (salary, performance), direct to HR contact"

Engineering Knowledge Base

Provides API documentation, architecture decisions, and coding standards to developers

Libraries: API Docs, Architecture Diagrams, Code Standards

Base Case: "Always include code examples and link to detailed documentation"

Sales Enablement

Assists sales team with product information, competitive analysis, and customer case studies

Libraries: Product Specs, Case Studies, Competitor Analysis

Base Case: "Emphasize unique selling points and customer success stories"

Best Practices

Create Focused Assistants

Instead of one general assistant, create multiple specialized assistants for different topics (HR, Engineering, Sales). This improves answer quality and allows targeted behavior customization.

Write Clear "Used For" Descriptions

When linking Libraries, clearly describe what information each Library contains. This helps the AI understand when to search each Library.

Test Base Cases Thoroughly

After adding base case instructions, test them with real questions to ensure the assistant behaves as expected. Refine instructions based on actual usage.

Keep Libraries Up-to-Date

Assistants are only as good as their linked Libraries. Use Crawlers to keep documentation synchronized and regularly review Library content for accuracy.