Product Overview
What is apptor flow?
apptor flow is a visual workflow automation platform built for teams that need to automate complex, multi-step business processes — especially those involving AI, human interaction, and third-party integrations.
Unlike low-code automation tools that are limited to simple sequences, apptor flow is built around a full execution engine. Workflows are first-class programs: they support branching, looping, parallel execution, error handling, retries, timeouts, subprocesses, and AI-native features like tool calling, memory, and streaming UI.
The Problem It Solves
Modern business processes require:
- Coordinating multiple systems (CRM, email, databases, AI models)
- Handling conditional logic and edge cases
- Involving humans at key decision points
- Running AI agents that call tools, remember context, and interact with users dynamically
Building these by hand requires significant engineering effort and ongoing maintenance. apptor flow provides a platform where these workflows can be designed visually, executed reliably, and monitored in real time — without needing a custom backend for each process.
Who Uses apptor flow
| Role | How They Use It |
|---|---|
| Product teams | Design AI-powered customer-facing workflows without writing backend code |
| Integration engineers | Connect internal systems using HTTP, SQL, and domain integrations |
| Operations teams | Automate repetitive multi-step processes and monitor them through the admin dashboard |
| Platform engineers | Extend the engine with custom node handlers, plugins, and vertical integrations |
| Developers | Execute workflows programmatically via the REST API using API Keys |
Core Capabilities
Visual Workflow Designer
A browser-based, drag-and-drop designer built on the GoJS diagram library. Users can:
- Add nodes from a categorized node palette
- Connect nodes with typed connections (sequence, success, error, timeout, conditional)
- Configure every node property in a live properties panel
- Switch between left-to-right and top-to-bottom layout
- Undo/redo, copy/paste, multi-select nodes
20+ Built-in Node Types
Organized into five categories:
| Category | Examples |
|---|---|
| Flow Control | Start Event, End Event, If/Else, Split, Loop, Call Process, Subprocess |
| AI | AI Task (LLM + tools), Voice Task, Memory Action, Tool nodes |
| Integration | REST API, Email, SMS, SQL, Domain Task |
| Logic | Script Task (JavaScript/Python), Set Variable |
| Human | Input Node (forms), Output Node (display) |
Actor-Based Execution Engine
Each node type runs in its own actor pool with configurable thread counts. The engine is:
- Asynchronous — nodes execute concurrently where the workflow allows
- Resilient — per-node timeout and retry configuration
- Observable — real-time log streaming via Server-Sent Events (SSE)
- Persistent — execution state stored in PostgreSQL, survivable across restarts
AI-Native Features
- AI Task — connects to any LLM provider, supports tool calling, streaming, memory, and A2UI
- A2UI — AI generates interactive UI components directly in chat (24 component types including tables, forms, buttons, charts)
- Voice Task — voice call automation with Twilio + OpenAI voice models
- Memory Action — manage conversation memory across workflow executions
- Tool nodes — expose email, SQL, REST, MCP, knowledge base, web search, and file operations as LLM tools
- RAG / Knowledge Base — vector store integration for retrieval-augmented generation
Multi-Tenant Architecture
- Organizations are isolated via
organizationIdon every data entity - Subdomain-based tenant resolution
- RBAC: roles, permissions, and user groups scoped to organizations
- Super Admin: cross-organization management
Publishing and Public Access
- Workflows can be published as public URLs (
/w/{urlSlug}) - Public workflows render as a form, chat interface, or custom layout
- No authentication required for public executions — the published workflow enforces its own access
Platform Technology
| Layer | Technology |
|---|---|
| Backend runtime | Java 21, Micronaut 4.4.2 |
| Execution engine | Custom actor framework (apptor-actor-framework) |
| AI integration | LangChain4j |
| Database | PostgreSQL 16, Liquibase migrations |
| Cache / distributed state | Hazelcast |
| Frontend | Angular 18+, TypeScript |
| Workflow designer | GoJS diagram library |
| Messaging | In-memory-distributed queues (extensible to Kafka/RabbitMQ) |
| Auth | OIDC (browser), API Key (M2M) |
What apptor flow is NOT
- It is not a simple IFTTT-style trigger-action tool — it is a full execution engine
- It is not an ETL pipeline tool — though ETL-style flows can be built on it
- It is not a chat platform — though it can power chat-based AI workflows via A2UI
- It is not a BPM system in the BPMN sense — though it shares BPM concepts
Next Steps
- Core Concepts → — understand workflows, nodes, variables, triggers, and actors
- Architecture Overview → — understand how the platform is built
- Getting Started → — build your first workflow