PebbleAgent - Visual AI Builder

Visual AI Builder Guide (Productivity)

The Visual AI Builder is PebbleAI’s powerful no-code platform for creating custom AI workflows. As part of our Productivity suite (powered by Flowise technology), it allows you to build sophisticated AI applications by simply dragging and dropping components.

Understanding the Visual Builder

Note: The Visual AI Builder is part of PebbleAI’s Productivity features. For monitoring and analysis of your AI applications, see our Observability Overview.

What Can You Build?

The Visual AI Builder enables you to create:

  • Chatbots: Customer service bots, FAQ assistants, lead qualification
  • Document Processors: PDF analyzers, report generators, content extractors
  • Data Pipelines: ETL workflows with AI transformation
  • Multi-Agent Systems: Complex workflows with multiple AI agents
  • API Integrations: Connect AI to your existing tools
  • Knowledge Bases: RAG (Retrieval Augmented Generation) systems

Key Concepts

Nodes

The building blocks of your AI workflows:

  • Input Nodes: Where data enters (chat, files, APIs)
  • Processing Nodes: AI models, transformations, logic
  • Output Nodes: Where results go (chat, database, webhook)

Connections

Links between nodes that define data flow:

  • Drag from output to input ports
  • Data types must match
  • Multiple connections possible

Chatflows vs Agentflows

  • Chatflows: Linear conversational AI (simpler, faster)
  • Agentflows: Complex multi-step agent orchestration

Getting Started

Accessing the Builder

  1. From your PebbleAI dashboard, click Chatflows or AgentFlows
  2. Click Add New to create a new workflow
  3. The visual canvas opens with a blank workspace

Your First Chatflow

Let’s build a simple AI assistant:

  1. Add a Chat Model

    • From the sidebar, find “Chat Models”
    • Drag “ChatOpenAI” onto the canvas
    • Configure with your API credentials

  2. Add Memory

    • Drag “Buffer Memory” from the Memory section
    • This helps the AI remember conversation context
    • Connect it to the Chat Model’s memory input

  3. Add Input/Output

    • The chat interface is automatically included
    • This handles user input and displays responses

  4. Test Your Flow

    • Click Save Chatflow
    • Click Test to open the chat interface
    • Start chatting with your AI!

Core Components

Language Models

Chat Models

  • OpenAI: GPT-4, GPT-3.5
  • Anthropic: Claude models
  • Google: Gemini models
  • Open Source: Llama, Mistral via providers

Configuration Options

  • Temperature (creativity level)
  • Max tokens (response length)
  • System prompts (behavior instructions)
  • Model selection

Memory Systems

Keep conversations coherent with memory:

  • Buffer Memory: Stores recent messages
  • Summary Memory: Summarizes long conversations
  • Entity Memory: Tracks mentioned entities
  • Conversation Memory: Full conversation history

Document Loaders

Import and process various file types:

  • PDF Loader: Extract text from PDFs
  • CSV Loader: Process spreadsheet data
  • Web Scraper: Pull content from websites
  • API Loader: Fetch data from APIs

Vector Stores

Enable semantic search over documents:

  • In-Memory: Quick testing
  • Pinecone: Scalable cloud solution
  • Chroma: Open-source option
  • Weaviate: Enterprise features

Tools & Functions

Extend AI capabilities:

  • Calculator: Mathematical operations
  • Web Search: Real-time information
  • API Calls: Custom integrations
  • Code Interpreter: Execute Python

Advanced Features

Prompt Engineering

Craft effective prompts:

You are a helpful customer service agent for {company_name}.
Always:
- Be polite and professional
- Provide accurate information
- Escalate when unsure

Customer Context: {customer_info}

Use variables for dynamic prompts:

  • {input} - User’s message
  • {history} - Conversation history
  • Custom variables from your flow

Conditional Logic

Add decision-making to your flows:

  1. If/Else Nodes: Route based on conditions
  2. Classifiers: AI-powered routing
  3. Validators: Check data before processing

Multi-Agent Orchestration

Build complex systems:

  1. Supervisor Agent: Coordinates other agents
  2. Specialist Agents: Handle specific tasks
  3. Tool Agents: Execute actions
  4. Review Agents: Quality control

RAG (Retrieval Augmented Generation)

Build knowledge-based systems:

  1. Document Upload: Import your knowledge base
  2. Chunking: Split into manageable pieces
  3. Embedding: Convert to vectors
  4. Retrieval: Find relevant information
  5. Generation: Create contextual responses

Best Practices

Design Principles

  1. Start Simple: Build basic flows first
  2. Test Often: Validate each component
  3. Document Well: Name nodes clearly
  4. Modular Design: Create reusable sub-flows

Performance Optimization

  1. Cache Results: Reuse expensive operations
  2. Limit Tokens: Control response lengths
  3. Choose Right Models: Balance cost/performance
  4. Batch Operations: Process multiple items together

Error Handling

  1. Fallback Paths: Handle failures gracefully
  2. Validation: Check inputs before processing
  3. Logging: Track flow execution
  4. User Feedback: Inform users of issues

Security Considerations

  1. API Key Management: Never expose keys
  2. Input Validation: Sanitize user inputs
  3. Access Control: Limit who can edit flows
  4. Data Privacy: Handle sensitive data carefully

Common Use Cases

Customer Support Bot

Components needed:

  • Chat model with friendly persona
  • FAQ knowledge base
  • Escalation to human agent
  • Conversation memory

Document Analysis Tool

Components needed:

  • PDF loader
  • Text splitter
  • Embedding model
  • Vector store
  • Query interface

Lead Qualification Assistant

Components needed:

  • Conversational flow
  • Data validation
  • CRM integration
  • Scoring logic
  • Notification system

Deployment Options

Internal Use

  • Share with team members
  • Embed in internal tools
  • Use for productivity

External Deployment

  • Embed on website
  • API endpoint
  • Widget integration
  • White-label options

Troubleshooting

Common Issues

“Model not responding”

  • Check API credentials
  • Verify model limits
  • Review error logs

“Memory not working”

  • Ensure memory is connected
  • Check memory size limits
  • Verify session handling

“Documents not loading”

  • Check file format support
  • Verify file size limits
  • Review parsing errors

Getting Help

  1. Built-in Help: Hover over nodes for tooltips
  2. Templates: Start from pre-built examples
  3. Community: Share and learn from others
  4. Support: Contact PebbleAI team

Next Steps

PebbleChat GuidePebbleObserve - AI Usage