Two types of AI agents
FlowX provides agents for different stages of your application lifecycle:Config-time agents
Help you build apps fasterAI assistants embedded in FlowX Designer that help developers and business analysts create processes, UIs, and business rules.Used by: Designers, developers, analysts
Business agents
Power your end-user experiencesAI agents that run inside your applications, interacting with customers and employees to automate tasks and provide intelligent assistance.Used by: Your customers and employees
Config-time agents
These agents are built into FlowX Designer to accelerate app development:Analyst
Build and optimize BPMN processes using AI to generate, validate, and enhance workflows
Assistant
Answer questions about FlowX platform and your organization based on documentation
Designer
Convert natural language prompts, Figma designs, or sketches into UI layouts
Developer
Generate business rules and code expressions using natural language
See config-time agents in action
See config-time agents in action
Learn more about config-time agents
Detailed documentation for each config-time agent
Business agents
Agents that power intelligent experiences for your end-users. You have two options:FlowX accelerators
Pre-built business agents created by FlowX, ready to deploy:- Industry-specific solutions for banking, insurance, and more
- Tested and optimized for common use cases
- Deploy immediately or customize to your needs
Agentic Apps Marketplace
Browse and install FlowX accelerators
Custom agents
Use Agent Builder to create AI agents tailored to your specific business processes:- Document processing - Extract data from forms, contracts, and statements
- Content generation - Create personalized communications and reports
- Decision support - Analyze data and provide recommendations
- Conversational AI - Build chat-based customer interactions
Agent Builder
Create custom AI agents for your specific business needs
Integrating agents into apps
Once you have business agents (custom or from marketplace), integrate them into your applications:Chat interface
Conversational UI for end-users
BPMN integration
Trigger agents from workflows
API access
Programmatic integration
Enterprise security
| Security feature | Description |
|---|---|
| Zero-trust model | All components authenticate each other, preventing unauthorized access |
| Input/output scanning | Detect prompt injection, ban sensitive substrings, sanitize outputs |
| Custom fine-tuned models | Purpose-built models with curated data for each use case |
| Human oversight | Human-in-the-loop controls for ethical decision-making |
Personal data protection
FlowX ensures personal data never reaches AI models by design through controlled data ingestion, deterministic workflows, and encryption options.Data ingestion paths
Personal data enters the platform through three channels:| Channel | Data handling |
|---|---|
| Forms / UIs | Input data mapped directly to the process data model |
| 3rd party APIs | API responses mapped to the process data model |
| Documents | Structured docs (Word, Excel) have fields mapped automatically; unstructured docs require OCR extraction with optional human verification |
For unstructured documents containing sensitive data, a human operator can verify extracted data accuracy before it’s mapped to the data model.
Data storage
All ingested data ends up in data models - JSON collections stored in FlowX databases:- Data models are only accessible to deterministic business process steps
- Process logic controls when, how, and which data is sent to AI steps
- This ensures no personal data is sent to AI by design
Protection options
| Option | Description |
|---|---|
| Database encryption | Encrypt data at rest using database capabilities (e.g., pgcrypto in PostgreSQL) - prevents administrator access to business data |
| Sensitive data masking | Mark specific data model fields as sensitive to hide them in UIs and console outputs |
| Custom guardrails | Additional guardrails can be enforced in AI workflows |
Hallucination prevention
| Technique | Purpose |
|---|---|
| RAG | Ground responses in reliable external data sources |
| Multi-agent architecture | Dedicated agents for specific tasks with specialized processing |
| Structured outputs | Enforce JSON outputs for consistent, parseable responses |
| Model evaluation | Continuous testing to find optimal models for each task |
| Prompt tuning | Task-specific prompts to enhance accuracy and relevance |