AI Agent

The AI Agent enables natural language pipeline creation and management through a chat interface.

Overview

Instead of manually configuring pipelines, describe what you want in plain English:

“Create a pipeline that syncs HubSpot contacts to BigQuery every 6 hours”

The agent will:

1. Identify the source and destination
2. Suggest configuration options
3. Create the pipeline with your approval

Using the Agent

1. Open the Chat Interface

   Click “AI Agent” in the sidebar.

2. Describe Your Pipeline

   Examples:

   • “I want to sync Kafka messages to BigQuery”
   • “Set up a daily export from HubSpot to our data warehouse”
   • “Create a pipeline for streaming orders to analytics”

3. Answer Clarifying Questions

   The agent may ask about:

   • Specific streams to sync
   • Authentication credentials
   • Schedule preferences

4. Review and Confirm

   Before creating, the agent shows:

   • Full configuration preview
   • Estimated sync frequency
   • Resource requirements

5. Pipeline Created

   The agent creates the pipeline (disabled by default) and provides a link to view it.

Capabilities

Pipeline Creation

User: Create a HubSpot contacts pipeline to BigQuery

Agent: I'll help you create that pipeline. I need a few details:
       1. Which HubSpot stream? (contacts, companies, deals)
       2. What's your BigQuery project ID and dataset?
       3. How often should it run?

User: Contacts, project my-analytics, dataset crm_data, every 6 hours

Agent: Perfect! Here's the pipeline configuration:
       - Source: HubSpot > contacts
       - Destination: BigQuery > my-analytics.crm_data
       - Schedule: Every 6 hours (0 */6 * * *)

       Should I create this pipeline?

Connector Discovery

User: What sources are available?

Agent: Bizon Platform supports these sources:
       - HubSpot (contacts, companies, deals, tickets)
       - Kafka (streaming topics)
       - Custom sources (build your own)
       ...

Pipeline Updates

User: Change the HubSpot pipeline to run every 4 hours

Agent: I've updated the schedule from "0 */6 * * *" to "0 */4 * * *".
       The pipeline will now run every 4 hours.

Agent Architecture

The agent uses LangGraph with:

• PostgreSQL Checkpointer - Conversation state persists across sessions
• SSE Streaming - Real-time token delivery for smooth chat experience
• Tool Calls - Structured actions for connector discovery and pipeline creation

Available Tools

Tool	Description
list_sources	Get available source connectors
list_destinations	Get available destinations
create_pipeline	Create a new pipeline
get_pipeline	Get pipeline details
update_pipeline	Modify existing pipeline

Sandbox Agent

A specialized agent for custom source development:

User: Help me build a custom source for the Notion API

Agent: I'll help you create a Notion source connector.
       Please provide the Notion API documentation URL.

User: https://developers.notion.com/reference

Agent: Analyzing the documentation...
       I've generated a custom source with:
       - Streams: pages, databases, users
       - Authentication: API key (Bearer token)

       [Shows generated code]

       Would you like to preview it with your credentials?

Configuration

LLM Provider

Set your preferred AI model:

# OpenAI
LLM_PROVIDER=openai
LLM_MODEL=gpt-4o-mini
OPENAI_API_KEY=sk-xxx

# Anthropic
LLM_PROVIDER=anthropic
LLM_MODEL=claude-3-sonnet
ANTHROPIC_API_KEY=sk-ant-xxx

Database

The agent shares the platform database for checkpointing:

DATABASE_URL=postgresql+asyncpg://user:pass@host:5432/bizon

Safety Features

• Pipelines created disabled - Review before enabling
• Confirmation required - Agent asks before making changes
• Audit trail - All agent actions are logged
• Credential handling - Sensitive values masked in responses

Conversation History

Sessions persist across browser refreshes:

• View past conversations
• Continue where you left off
• Reference previous pipeline configurations