Skip to main content

A System is a complete workflow composed of one or more connected AI prompts that work together to process data sequentially. Think of a System as a recipe or blueprint that defines how information flows through different AI processing steps to achieve a specific outcome.

For example, a "Content Generation System" might include prompts for research, writing, and editing that process data step by step to create high-quality content. A "Customer Support System" might classify support tickets, generate responses, and determine escalation needs through multiple AI analysis steps.

Key Characteristics:​

  • Sequential Processing: Data flows from one AI node to the next in a defined order
  • Reusable Workflows: Once created, a System can be executed multiple times with different inputs
  • Configurable: Each System has its own settings for AI models, resource management, and processing parameters
  • Template-Based: Systems can be created from predefined templates or built from scratch

System Structure and Components​

Core System Properties​

Basic Information​

  • Name (string, required): A descriptive name for your System
  • Description (string, required): A detailed explanation of what the System does and its purpose
  • ID (string, auto-generated): Unique identifier for the System
  • Created At (Date): Timestamp when the System was first created
  • Updated At (Date): Timestamp when the System was last modified

Classification​

  • Tags (string[]): Array of tags for categorizing and organizing Systems
  • Is Template (boolean): Whether this System serves as a template for creating other Systems

Workflow Structure​

Every System contains a workflow that defines the processing logic:

Nodes​

Systems are composed of three types of nodes:

  1. Start Node: The entry point where input data enters the workflow
  2. AI Nodes: Processing units that contain AI prompts and configurations
  3. End Node: The exit point where final results are produced

AI Node Configuration​

Each AI Node has the following properties:

  • Label (string): Display name for the node
  • Prompt (string): The AI prompt/instruction text
  • Model (string): AI model to use (GPT-4o, GPT-4o-mini, GPT-5, etc.)
  • Temperature (number, 0-1): Controls creativity/randomness of AI responses
  • Max Tokens (number): Maximum number of tokens the AI can generate
  • System Prompt (string, optional): Additional context or instructions for the AI
  • Output Format (object): Defines how the AI should structure its response
    • Type: "text", "json", or "jsonSchema"
    • JSON Schema (optional): For structured data output
    • Relaxed (boolean, optional): Whether to use relaxed schema validation

Edges​

Connections between nodes that define the data flow path through the workflow.

Resource Settings​

Systems include optional Resource Settings that configure how the System handles knowledge bases and embeddings:

Embedding Configuration​

  • Embedding Enabled (boolean): Whether to use text embeddings for the System
  • Chunk Size (number): Size of text chunks in tokens (100-8000)
  • Chunk Overlap (number): Percentage overlap between chunks (0-50%)

Knowledge Base Configuration​

  • Knowledge Base Type (string): How to select relevant knowledge
    • "chunk_similarity_search": Find and use only the most relevant chunks
    • Other types (currently disabled): Document similarity, full dump, AI pre-filter, etc.
  • Number of Results (number): How many knowledge chunks to include (1-20)
  • Metric Type (string): Distance calculation method ("IP" for Inner Product or "L2" for Euclidean Distance)
  • Top K (number): Number of top matches to consider
  • N Probe (number): Search precision parameter (1-200)

How to Create a System​

Step 1: Access System Creation​

[Screenshot placeholder: Systems list page with "Create System" button]

  1. Navigate to the Systems page in the application
  2. Click the "Create System" button in the top-right corner

Step 2: Choose a Template​

[Screenshot placeholder: System template selection modal]

The template selection modal will appear with several options:

Available Templates:​

  1. Customer Support Automation

    • Automatically classifies support tickets
    • Generates responses using company knowledge
    • Determines escalation needs
    • Pre-configured with 3 AI nodes: Classification → Response Generation → Escalation Assessment

  2. Marketing Content Generation

    • Creates comprehensive marketing materials
    • Includes research, outline creation, content writing, and SEO optimization
    • Pre-configured with 4 AI nodes: Research → Outline → Content → SEO Optimization

  3. Document Processing (Structured Data Extraction)

    • Extracts structured data from documents
    • Validates information accuracy
    • Formats output into standardized schemas
    • Pre-configured with 3 AI nodes: Extraction → Validation → Schema Formatting

  4. Blank System

    • Empty workflow with just Start and End nodes
    • Perfect for building custom workflows from scratch

Step 3: Configure System Details​

[Screenshot placeholder: System name input field]

  1. System Name: Enter a descriptive name for your System

    • The template name will be pre-filled if you selected a template
    • You can modify this to suit your needs

  2. Click "Create System" to proceed

Step 4: Configure the Workflow​

[Screenshot placeholder: System Editor with workflow canvas]

After creation, you'll be taken to the System Editor where you can:

Edit Existing Nodes​

  • Click on any AI node to edit its configuration
  • Modify the prompt, model selection, temperature, and other parameters
  • Configure output format (text, JSON, or structured schema)

Add New Nodes​

  • Use the sidebar panel to add new AI nodes to your workflow
  • Choose from pre-built templates like Text Analyzer, Content Generator, Code Reviewer, etc.
  • Connect nodes by dragging from output handles to input handles

Workflow Navigation​

The editor provides:

  • Visual Canvas: Drag-and-drop interface for arranging nodes
  • Connection Lines: Visual representation of data flow between nodes
  • Edit Panel: Detailed configuration for selected nodes
  • Execution Controls: Run the entire workflow or individual nodes

Step 5: Configure Resource Settings (Optional)​

[Screenshot placeholder: Resources Settings tab]

Navigate to the Resources tab to configure:

Embedding Settings​

  • Enable/Disable embeddings for the System
  • Chunk Size: Adjust how text is split for processing (default: 1024 tokens)
  • Chunk Overlap: Set overlap between chunks (default: 15%)

Knowledge Base Selection​

  • Selection Type: Choose how relevant knowledge is selected
  • Number of Results: Set how many knowledge pieces to include (default: 10)
  • Advanced Settings: Configure metric types and search parameters

Step 6: Test and Save​

[Screenshot placeholder: Workflow execution with results panel]

  1. Test Your System: Use the execution controls to test your workflow
  2. Review Results: Check the output in the results panel
  3. Iterate: Modify prompts and settings based on test results
  4. Auto-Save: Changes are automatically saved as you work

System Management​

Viewing Systems​

[Screenshot placeholder: Systems list with search and filters]

The Systems list page provides:

  • Search Functionality: Find Systems by name or description
  • System Cards: Display name, last updated date, and step count
  • Quick Actions: Edit, duplicate, or delete Systems directly from the list

System Operations​

Editing​

  • Click on any System card or use the edit menu option
  • Opens the System Editor for full workflow modification

Duplicating​

  • Creates a copy of the existing System
  • Useful for creating variations or backups
  • New System will have "Copy of [Original Name]" as the default name

Deleting​

  • Permanently removes the System from your account
  • Warning: This action cannot be undone

Advanced Features​

Output Formatting​

Systems support sophisticated output formatting:

Text Output​

  • Simple text responses from AI nodes
  • Best for general content generation and analysis

JSON Output​

  • Structured data output
  • Useful for extracting specific information
  • Can include custom field definitions

JSON Schema Output​

  • Highly structured output with validation
  • Define exact fields, types, and requirements
  • Supports nested objects and arrays
  • Includes relaxed mode for flexible validation

Workflow Execution​

Execution Order​

  • Systems automatically calculate the correct execution order using topological sorting
  • Handles complex dependency relationships
  • Detects and prevents circular dependencies

Progress Tracking​

  • Real-time execution status for each node
  • Performance metrics including execution time
  • Error handling with detailed error messages
  • Comprehensive execution logs

Result Management​

  • Node outputs are cached for efficiency
  • Results can be viewed and exported
  • Failed executions show detailed error information

Best Practices​

System Design​

  1. Clear Naming: Use descriptive names that indicate the System's purpose
  2. Logical Flow: Arrange nodes in a logical sequence that makes sense for your use case
  3. Appropriate Models: Choose AI models based on task complexity and required accuracy
  4. Temperature Settings: Use lower temperatures (0.1-0.3) for factual tasks, higher (0.7-0.9) for creative tasks

Prompt Engineering​

  1. Clear Instructions: Write specific, clear prompts that explain exactly what you want
  2. Context Provision: Include relevant context and examples in your prompts
  3. Output Specification: Clearly specify the desired output format and structure
  4. Iterative Refinement: Test and refine prompts based on actual results

Resource Management​

  1. Embedding Usage: Enable embeddings when working with knowledge bases
  2. Chunk Size Optimization: Adjust chunk sizes based on your content type
  3. Result Limits: Set appropriate limits for knowledge base results to balance relevance and performance

Integration with Assistants​

Important Note: Systems by themselves cannot be executed directly by end users. They serve as templates or blueprints that must be converted into Assistants to become functional.

Systems vs. Assistants​

  • Systems: The workflow definition and configuration
  • Assistants: Executable instances of Systems with additional capabilities
  • Conversion: Systems are converted to Assistants to enable user interaction

Future Assistant Capabilities​

In upcoming versions, Assistants created from Systems will support:

  • RAG (Retrieval-Augmented Generation): Integration with document knowledge bases
  • Persistent Chat: Conversational interfaces with memory
  • API Integration: Connection to external services and databases
  • User Management: Access controls and user-specific configurations
  • Document Upload: Direct file upload and processing capabilities

Workflow​

  1. Create System: Design and configure your workflow using the System Editor
  2. Test System: Verify the workflow works as expected
  3. Create Assistant: Convert the System into an Assistant for user interaction
  4. Deploy: Make the Assistant available to end users

This documentation covers the complete lifecycle of Systems, from creation to integration with Assistants. Systems provide the foundation for building powerful AI workflows that can be executed and shared through the Assistant interface.