Managing Assistants

An Assistant is an executable instance of a System that enables end-user interaction and provides enhanced capabilities beyond the base System workflow. While Systems are the blueprint or template that defines AI processing workflows, Assistants are the functional, deployable versions that users can actually interact with.

Think of an Assistant as a specialized AI agent built from a System that can:

Chat with users in real-time with conversational memory
Access knowledge bases through uploaded documents and files
Execute System workflows with user-provided inputs
Maintain session history for tracking interactions and results
Provide streaming responses for real-time communication

Key Characteristics:

System-Based: Every Assistant is built from an existing System's workflow
Interactive: Users can chat directly with Assistants through a conversational interface
Knowledge-Enhanced: Support for RAG (Retrieval-Augmented Generation) with document upload
Persistent: Maintains conversation history and session data
Configurable: Each Assistant has its own settings separate from the underlying System

Assistant Structure and Components

Core Assistant Properties

Basic Information

Name (string, required): A descriptive name for your Assistant
Description (string, required): Explanation of what the Assistant does and its purpose
ID (string, auto-generated): Unique identifier for the Assistant
System ID (string, required): Reference to the System this Assistant is based on
Created At (Date): Timestamp when the Assistant was first created
Updated At (Date): Timestamp when the Assistant was last modified

Classification and Access

Tags (string[]): Array of tags for categorizing and organizing Assistants
Is Public (boolean): Whether this Assistant is publicly accessible or private

Assistant Configuration

Every Assistant includes a comprehensive configuration object that controls its behavior:

Core Settings

Temperature (number, 0-1): Controls creativity/randomness of AI responses (inherited from System)
Max Tokens (number): Maximum number of tokens the Assistant can generate per response (inherited from System)
Enable RAG (boolean): Whether to use uploaded knowledge base documents for enhanced responses

RAG (Retrieval-Augmented Generation) Settings

Enable RAG (boolean): Whether to use knowledge base documents for enhanced responses
RAG Settings (object, optional):
- Enabled (boolean): Master toggle for RAG functionality
- Document Sources (string[]): List of knowledge base document identifiers
- Chunk Size (number): Size of text chunks for processing
- Overlap Size (number): Overlap between text chunks
- Similarity Threshold (number): Minimum similarity score for including knowledge chunks

Chat Settings

Chat Settings (object, optional):
- Enabled (boolean): Whether chat functionality is active
- Max History (number): Maximum number of previous messages to remember
- Context Window (number): Total token limit for conversation context
- Personality Prompt (string, optional): Custom personality instructions

How to Create an Assistant

Step 1: Access Assistant Creation

[Screenshot placeholder: Assistants list page with "Create Assistant" button]

Navigate to the Assistants page in the application
Click the "Create Assistant" button in the top-right corner

Step 2: Configure Basic Information

[Screenshot placeholder: Assistant creation form]

The Assistant creation form includes:

Assistant Details

Assistant Name: Enter a descriptive name (e.g., "Marketing Helper", "Code Reviewer")
System Selection: Choose the System this Assistant will be based on
- Dropdown shows all available Systems with descriptions
- The Assistant will inherit the System's workflow and AI model configuration

Step 3: Automatic Configuration

When you create an Assistant, it automatically inherits the AI model configuration from the selected System and applies default Assistant-specific settings:

{
  "enableRag": false,
  "chatSettings": {
    "enabled": true,
    "maxHistory": 10,
    "contextWindow": 4000
  }
}

The AI model, temperature, max tokens, and system prompts are inherited directly from the underlying System's workflow nodes.

Step 4: Assistant Management

After creation, you're automatically taken to the Assistant Management interface with three main tabs:

Assistant Management Interface

Knowledge Base Tab

[Screenshot placeholder: Knowledge base management interface]

The Knowledge Base tab allows you to enhance your Assistant with documents and files:

File Upload

Drag & Drop: Drag files anywhere on the tab to upload them
File Browser: Click "Add Files" to select files from your computer
Supported Formats: PDF, DOC/DOCX, TXT, MD, CSV, JSON, HTML, and more

Upload Process

The system handles file processing in three stages:

Getting Upload URL: Secure signed URL generation for S3 storage
Uploading to S3: Direct file upload to cloud storage
Processing: Text extraction, chunking, and embedding generation

File Management

File Status: Monitor processing status (Available, Extracting Text, Embedding, etc.)
File Details: View size, type, upload date, and extracted text
File Actions: Inspect content, download original files, or remove from knowledge base

Knowledge Base Features

Search: Find specific files in your knowledge base
Status Monitoring: Real-time updates on file processing progress
Text Extraction: Automatic text extraction from various document types
Embedding Generation: AI-powered text embeddings for similarity search

Playground Tab

[Screenshot placeholder: Chat playground interface]

The Playground provides a real-time chat interface for testing and interacting with your Assistant:

Chat Interface

Conversational UI: Natural chat experience with message history
Streaming Responses: Real-time response generation with live updates
Message History: Persistent conversation within the session
Rich Formatting: Support for markdown formatting in responses

Chat Features

Knowledge Integration: Automatic use of uploaded documents when relevant
Debug Information: Access to detailed execution logs for each response
Response Streaming: Watch responses generate in real-time
Error Handling: Clear error messages for failed interactions

User Experience

Auto-scroll: Automatic scrolling to new messages
Typing Indicators: Shows when the Assistant is processing
Send Methods: Enter to send, Shift+Enter for new lines
Response Status: Visual indicators for message processing states

History Tab

[Screenshot placeholder: Session history interface]

The History tab provides comprehensive tracking of all Assistant interactions:

Session Management

Session List: All previous conversations with timestamps
Search Functionality: Find specific sessions by input or output content
Session Details: Input messages, AI responses, and execution metadata
Chronological Order: Sessions sorted by most recent first

Data Tracking

Input/Output Pairs: Complete record of user inputs and Assistant responses
Timestamps: Precise timing information for each interaction
Session Metadata: Additional data about execution context
Export Capabilities: Access to session data for analysis

Advanced Features

Real-Time Streaming

Assistants support advanced streaming capabilities:

Server-Sent Events (SSE)

Live Response Generation: Responses appear as they're generated
Progress Indicators: Real-time status updates during processing
Error Recovery: Graceful handling of connection issues
Multi-step Processing: Status updates for complex System workflows

Streaming Benefits

Immediate Feedback: Users see responses starting immediately
Better UX: No waiting for complete responses
Progress Transparency: Clear indication of processing stages
Reduced Latency: Perceived faster response times

Knowledge Base Integration

RAG (Retrieval-Augmented Generation)

Automatic Context: Relevant documents automatically included in responses
Similarity Search: AI-powered matching of user queries to knowledge content
Chunk Management: Intelligent text chunking for optimal retrieval
Context Windows: Balanced inclusion of relevant knowledge without overwhelming

Document Processing Pipeline

Upload: Secure file upload to cloud storage
Text Extraction: Convert documents to searchable text
Chunking: Split content into manageable pieces
Embedding: Generate AI embeddings for similarity search
Indexing: Make content searchable for real-time retrieval

Session Management

Conversation Persistence

Cross-Session Memory: Conversations persist across browser sessions
History Tracking: Complete record of all interactions
Context Preservation: Maintain conversation context within sessions
Data Analytics: Session data available for usage analysis

Session Features

Automatic Saving: All interactions automatically saved
Search Capabilities: Find previous conversations quickly
Export Options: Access to session data in structured formats
Privacy Controls: Manage data retention and access

Assistant Operations