> For the complete documentation index, see [llms.txt](https://whitepaper.aitech.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://whitepaper.aitech.io/agentforge/connections/basics.md).

# basics

When you run a workflow in Agent Forge, the execution engine follows a systematic process to ensure blocks are executed in the correct order with proper data flow.

### Execution Flow

The execution engine runs workflows in layers, processing blocks based on their dependencies:

{% stepper %}
{% step %}

### Validation

Ensures the workflow has a starter block with no incoming connections and all blocks are properly connected.
{% endstep %}

{% step %}

### Layer-based Execution

Identifies which blocks can execute next based on completed dependencies and executes them in parallel.
{% endstep %}

{% step %}

### Path Updates

Router and Condition blocks update the active execution path, determining which blocks execute next.
{% endstep %}

{% step %}

### Iteration Processing

Loop and Parallel blocks manage iterations and create virtual instances for concurrent execution.
{% endstep %}

{% step %}

### Result Collection

Outputs from the final blocks are collected and returned as the workflow result.
{% endstep %}
{% endstepper %}

### Block Types and Execution

Different block types have different execution behaviors:

{% tabs %}
{% tab title="Orchestration Blocks" %}
Orchestration blocks control the flow of execution through your workflow.

* **Starter Block**: Initiates workflow execution and provides initial input values. Every workflow must have exactly one starter block.
* **Router Block**: Directs execution along specific paths based on dynamic decisions. Uses an AI model to select one of multiple possible paths.
* **Condition Block**: Executes different paths based on conditional logic. Evaluates JavaScript expressions to determine which path to follow.
* **Loop Block**: Executes blocks repeatedly for a fixed number of iterations or over a collection. Manages iteration state and provides access to current item.
* **Parallel Block**: Executes blocks concurrently across multiple instances. Distributes work based on count or collection for faster processing.
  {% endtab %}

{% tab title="Processing Blocks" %}
Processing blocks transform data and generate new outputs.

* **Agent Block**: Interacts with AI models to generate content. Executes prompts against various LLM providers.
* **Function Block**: Executes custom JavaScript/TypeScript code. Runs in a secure sandbox environment with access to connected block outputs.
* **Evaluator Block**: Assesses outputs against defined criteria. Uses AI to evaluate content based on custom metrics.
* **Workflow Block**: Execute nested workflows as a single block. Allows modular workflow design by embedding one workflow inside another.
  {% endtab %}

{% tab title="Integration Blocks" %}
Integration blocks connect with external systems.

* **API Block**: Makes HTTP requests to external services. Configurable with headers, body, and authentication.
* **Knowledge Base Block**: Search and interact with knowledge bases. Performs semantic search, retrieves documents, and manages knowledge data.
* **Response Block**: Format and return responses from workflows. Configure response data, status codes, and headers.
* **Tool Blocks**: Specialized blocks for specific services (Gmail, Slack, GitHub, etc.). Each has its own execution logic for the specific service.
  {% endtab %}
  {% endtabs %}

### Execution Methods

Agent Forge offers multiple ways to trigger workflow execution:

#### Manual Execution

Run workflows on-demand through the Agent Forge interface by clicking the "Run" button. This is perfect for:

* Testing during development
* One-off tasks
* Workflows that need human supervision

#### Scheduled Execution

Configure workflows to run automatically on a specified schedule:

* Set up recurring executions using cron expressions
* Define start times and frequency
* Configure timezone settings
* Set minimum and maximum execution intervals

#### API Endpoints

Each workflow can be exposed as an API endpoint:

* Get a unique URL for your workflow
* Configure authentication requirements
* Send custom inputs via POST requests
* Receive execution results as JSON responses

**Viewing Deployed APIs**

Monitor your deployed workflow APIs and their current state:

This shows how to view the deployed state and compare with the original deployed API configuration.

#### Webhooks

Configure workflows to execute in response to external events:

* Set up webhook triggers from third-party services
* Process incoming webhook data as workflow input
* Configure webhook security settings
* Support for specialized webhooks (GitHub, Stripe, etc.)

{% hint style="info" %}
The execution method you choose depends on your workflow's purpose. Manual execution is great for development, while scheduled execution, API endpoints, and webhooks are better for production use cases.
{% endhint %}

### Execution Context

Each workflow execution maintains a context that tracks:

* **Block States**: Outputs and execution status of each block
* **Execution Path**: Active blocks based on routing decisions
* **Loop/Parallel State**: Current iterations and distribution items
* **Environment Variables**: Configuration values available during execution
* **Execution Logs**: Detailed records of each block's execution

### Real-Time Monitoring

Monitor your workflow execution in real-time:

* **Active Block Highlighting**: Currently executing blocks pulse with animation
* **Live Logs**: Execution logs appear instantly in the logs panel
* **Block States**: Visual indicators show success, error, or pending states
* **Performance Metrics**: Execution time for each block

These monitoring features help you understand workflow behavior and quickly identify any issues.