Architecture
Automation Hub is designed to be simple for users while being powerful under the hood. This page explains the key components and how they work together to run your automations reliably at scale.
How It Works
Every automation follows this path from idea to result:
Key Components
Web Interface & Canvas
What you see and use every day.
The Canvas is your visual workspace for building automations. It is an Angular-based web application where you drag and drop steps onto a workflow designer, connect them, and configure each one.
| Capability | What it does |
|---|---|
| Workspace & Collection management | Organize automations by team, project, or purpose |
| Visual workflow designer | Build workflows by dragging steps onto a canvas and connecting them |
| Card & Grid views | Browse your automations in the layout that works for you |
| Search | Find workspaces, collections, jobs, and steps across your environment |
| Execution monitoring | Watch runs in real time with step-level status and output |
| Import & Export | Move automations between environments or share with other teams |
AI Integration (MCP Server)
Let AI agents work with your automations.
The MCP Server is a dedicated service that speaks the Model Context Protocol, an open standard for AI agent communication. It exposes 9 tool methods that allow AI agents like Claude, OpenAI, and others to discover, run, and monitor your automations through natural language.
An AI agent can ask: "Run the nightly data sync job" or "Show me the status of the last invoice processing run" and the MCP Server translates that into the right platform actions.
| What AI agents can do | How it works |
|---|---|
| Discover automations | Search and list available jobs across workspaces |
| Execute workflows | Trigger runs with parameters, just like a manual run |
| Monitor progress | Check execution status, step outputs, and errors |
| Review history | Look at past runs and their results |
The MCP Server is stateless, meaning it does not store session data. Every request is independent, which makes it reliable and easy to scale.
Execution Engine (Job Executor)
The brain that runs everything.
When you trigger an automation, the Execution Engine takes your workflow definition and turns it into an execution plan. Rather than running steps one at a time in a straight line, the engine analyzes the dependencies between your steps and builds a Directed Acyclic Graph (DAG) - a smart map of what needs to happen and in what order.
The engine then creates a durable workflow using Temporal, an enterprise-grade orchestration system. This means your automations survive server restarts, handle retries automatically, and can be paused, resumed, or stopped at any point.
| Capability | What it means for you |
|---|---|
| DAG execution planning | Steps that do not depend on each other run at the same time |
| Durable workflows | Runs survive server restarts without losing progress |
| Pause, resume, and stop | Full control over running automations |
| Conditional branching | Steps can follow different paths based on data |
| Human-in-the-loop approvals | Pause for human review before continuing |
| Automatic retries | Recover from temporary failures without manual intervention |
Plugin System
One plugin for every type of action.
Automation Hub has 30+ plugins organized by category. Each plugin knows how to perform one specific type of action: send an email, call an API, query a database, run an SSH command, and so on.
| Category | Examples |
|---|---|
| Communication | Email, Slack, Microsoft Teams |
| Cloud | AWS S3 |
| Enterprise Apps | Salesforce |
| Protocol | SSH, FTP, SFTP, FTPS |
| API | REST API |
| Database | SQL databases, NoSQL databases |
| Data | JSON, CSV, XML processing |
| Utility | Variables, formatting, date-time, cryptography |
| AI | AI Agent (LLM integration) |
| Flow control | Conditional, Loop, Group, Merge, Split |
Plugins are discovered automatically and loaded on demand. When your workflow reaches a step that sends an email, the platform loads the email plugin, runs it, and moves on. You never have to install or manage plugins yourself.
Connectors
The bridge between plugins and external systems.
Behind every plugin is one or more connectors. Connectors handle the low-level details of communicating with external systems: managing network connections, handling authentication protocols, translating data formats, and dealing with protocol-specific quirks.
There are 27 connector modules covering SMTP, SSH, HTTP, JDBC, SFTP, FTP, Kafka, Salesforce, Microsoft APIs, Slack, AWS S3, and more. When a plugin needs to send an email, it delegates to the email connector, which handles the SMTP handshake, authentication, and delivery. This separation means plugins stay simple and focused on business logic.
Global Configurator
Keeps sensitive data out of your workflows.
The Global Configurator is a centralized service for managing everything your automations need to connect to external systems.
| What it manages | Why it matters |
|---|---|
| Connections | Credentials, hosts, ports, tokens, and API keys for external systems |
| Global variables | Shared values that any automation can reference |
| Datasets | Reusable data collections for automations |
| Files | Shared files like templates, certificates, or reference data |
| Certificates | TLS/SSL certificates for secure connections |
Instead of storing a database password inside every workflow that needs it, you create a connection once in the Global Configurator and reference it by name. When credentials change, you update them in one place.
Execution Archival Service
Keeps your system fast without losing history.
Over time, execution records accumulate. The Archival Service automatically moves old execution records from the active database to cold storage on a configurable schedule.
| Setting | What it controls |
|---|---|
| Time-window mode | Archive executions older than a specified number of days |
| Per-job retention | Keep a set number of recent executions per job, archive the rest |
| Schedule | How often the archival process runs |
| Storage destination | Where archived records go (S3 or MinIO compatible storage) |
Archived records are not deleted. They are moved to cheaper storage so the active database stays fast and responsive.
How Your Automation Actually Runs
Here is the real path a request takes through the platform:
In more detail:
- A trigger fires - You click Run, a schedule fires, a webhook arrives, an API call comes in, or an AI agent makes a request through the MCP Server.
- The Execution Engine receives the request - It loads the job definition, validates connections and inputs, and resolves any variables from the Global Configurator.
- The engine builds a DAG - It analyzes the dependencies between your steps and creates an execution plan. Steps that can run at the same time are grouped together.
- A durable workflow is created - The engine hands the plan to Temporal, which ensures the run will complete even if a server restarts mid-execution.
- Steps execute - For each step, the engine loads the right plugin, the plugin delegates to the right connector, and the connector communicates with the external system. Outputs from each step are passed to the next steps that need them.
- Results are recorded - Every step's status, output, duration, and any errors are saved. The overall run status is updated in real time.
Smart Execution: The DAG Engine
Most workflow tools run steps one after another in a straight line. Automation Hub is smarter than that.
What Is a DAG?
A DAG (Directed Acyclic Graph) is the execution plan that the engine builds from your workflow. It is a map of your steps and their dependencies - which steps need data from which other steps, and which steps can run independently.
The key rules:
- Directed means data flows in one direction, from earlier steps to later steps
- Acyclic means there are no circular dependencies (Step A cannot depend on Step B if Step B already depends on Step A)
- The engine validates these rules when you save your workflow, so problems are caught before you run anything
Parallel Execution
When the engine builds the DAG, it identifies steps that do not depend on each other and runs them at the same time.
Example: Imagine a workflow that needs to fetch data from three different systems and then combine the results.
After all three complete, a fourth step combines the results. Without parallel execution, these three fetches would run one after another, taking three times as long.
Conditional Branching
Steps can follow different paths based on data from earlier steps.
When a condition is evaluated, only the matching branch runs. Steps on the other branch are skipped and marked accordingly in the execution record.
What This Means for You
| Benefit | How the DAG engine delivers it |
|---|---|
| Faster runs | Independent steps run in parallel instead of waiting in line |
| Smarter workflows | Conditional branching lets one workflow handle multiple scenarios |
| Early error detection | The engine validates your workflow structure before it runs |
| Clear execution history | Every step shows whether it ran, was skipped, or is waiting |
Where Your Data Lives
Automation Hub uses three storage systems, each chosen for a specific purpose.
| Storage | What it holds | Why this system |
|---|---|---|
| MongoDB | Jobs, steps, execution records, connections, configurations | Flexible document storage that handles the varied shapes of automation data |
| Temporal | Active workflow state, retry queues, task scheduling | Purpose-built for durable workflow orchestration with built-in fault tolerance |
| S3 / MinIO | Archived execution records, large outputs | Cost-effective storage for historical data that is accessed infrequently |
How They Work Together
- MongoDB is the primary database. When you create a job, add steps, save a connection, or run an automation, the data goes here. It is fast and always available.
- Temporal manages running workflows. When an automation is executing, Temporal tracks which step is active, handles retries, manages timeouts, and ensures the run completes even if a server restarts. Once the run finishes, the results are written back to MongoDB.
- S3 / MinIO is cold storage. The Archival Service periodically moves old execution records from MongoDB to S3-compatible storage. This keeps MongoDB fast while preserving your complete execution history.
Why This Architecture Works
Simple for Users
- Focus on what you want to automate, not how the system works internally
- Visual building blocks instead of programming
- AI agents can run your automations using natural language
- One place for all your connections and credentials
Powerful Under the Hood
- DAG engine runs independent steps in parallel for faster execution
- Durable workflows survive server restarts without losing progress
- Handles thousands of concurrent automations
- Scales each component independently as your needs grow
Secure by Design
- Credentials are stored centrally in the Global Configurator, not in workflows
- Encrypted connections to external systems
- Role-based access control across workspaces
- Complete audit trails for every automation run
Enterprise Ready
- Temporal orchestration provides the reliability that production workloads demand
- Automatic archival keeps the system performant over time
- Supports compliance and governance requirements with full execution history
- High availability and disaster recovery through proven infrastructure components
Related Pages
- Jobs and Steps - understand the building blocks of a workflow
- Execution Lifecycle - follow a run from trigger to result
- Workspaces and Collections - organize your automations
- Integration Nodes - browse the full catalog of available nodes
- AI Integration - connect AI agents to your automations