Skip to main content
Patterns are proven solutions to common challenges when building AI apps on FlowX. Each pattern is extracted from production implementations and can be combined to build complex apps.

​
Available patterns

Agent typologies

The nine recurring agent archetypes behind the FlowX.AI agent catalog, mapped to the nodes that implement them. Start here to pick a shape.

Intent classification and routing

Use a TEXT_UNDERSTANDING node to classify user input and route to specialized handlers. The foundation of any conversational AI app.

Knowledge base RAG

Ground AI responses in your documents using retrieval-augmented generation with Qdrant vector search and the CUSTOM_AGENT node.

Fan-out extraction

Classify documents by type, then route each to a specialized TEXT_EXTRACTION node with tailored prompts and schemas. Scale to dozens of document types.

AI comparison and reconciliation

Compare AI-extracted document data against system-of-record values and generate structured exception reports with match rates and confidence scores.

Hybrid AI + business rules

Combine AI extraction and understanding with deterministic business logic (formulas, eligibility checks, scoring) for auditable decision-making.

Session state management

Manage conversation history and session state across multi-turn interactions using FlowX Database workflows.

​
How to use patterns

Each pattern page includes:
  • When to use — the problem it solves and when to reach for it
  • Architecture — how the workflow nodes connect
  • Implementation — key configuration: prompts, schemas, fork conditions
  • Real-world example — where this pattern appears in the tutorials
  • Variations — common adaptations
Patterns are building blocks. A typical AI app combines 2-4 patterns:

​
Multi-agent orchestration

Combining patterns produces multi-agent apps: several specialized agents coordinated inside one workflow, each with its own prompt, schema, and model.
  • Router and specialists: an intent classification agent reads the input and dispatches to specialized handler agents — the foundation of conversational apps.
  • Parallel specialists: fan-out extraction classifies documents by type and routes each to a dedicated extraction agent with a tailored prompt and schema, scaling to dozens of document types processed in parallel.
  • Agents plus deterministic logic: hybrid AI + business rules places auditable business logic and approval gates between agent steps, so multi-agent output stays governed.
Each agent in the composition can use a different model, and the whole orchestration is testable with evaluations per AI node.

​
Pattern origins

These patterns are extracted from two production-grade FlowX apps:
  • Mortgage advisor — a chatbot that evaluates loan eligibility across 7 banks using conversational AI, document extraction, and financial calculations
  • Logistics document processor — an email-triggered pipeline that processes 17 document types from carrier emails, reconciles them against a TMS, and surfaces exceptions for review
See the Tutorials for full implementations.

Tutorials

End-to-end implementations using these patterns

Node types

AI node type reference

Chat-driven workflows

Multi-turn chat with session memory

Using agents

Integration and deployment options
Last modified on September 10, 2026