Skip to main content

​
Overview

Actions are the building blocks of pipelines. Each action performs a specific operation and can be configured with parameters, outputs, and conditions.

​
Control Flow Actions

​
Log

Write messages to execution logs for debugging and monitoring. Parameters:
  • message (string, required): The message to log
  • level (string): Log level - info, warning, error, debug
Example: 
{
    "action": "log",
    "params": {
        "message": "Starting deployment process",
        "level": "info"
    }
}

​
Conditional Branch

Execute different paths based on conditions. Parameters:
  • condition (string, required): Boolean expression to evaluate
  • branches (object): Named branches with their conditions
Example: 
{
    "action": "conditional",
    "params": {
        "condition": "steps.health_check.output.status == 'healthy'"
    }
}

​
Transform

Modify or transform data between steps. Parameters:
  • operation (string, required): Transformation type
  • input (any): Input data to transform
  • template (string): Jinja2 template for transformation

​
Aggregate

Combine outputs from multiple parallel actions. Parameters:
  • sources (array, required): List of step IDs to aggregate
  • strategy (string): Aggregation strategy - merge, concat, first

​
Set Output

Define pipeline-level outputs. Parameters:
  • key (string, required): Output variable name
  • value (any, required): Output value
Example: 
{
    "action": "set_output",
    "params": {
        "key": "endpoint_url",
        "value": "{{steps.deploy.output.endpoint_url}}"
    }
}

​
Wait Until

Pause execution until a condition is met. Parameters:
  • condition (string, required): Condition to wait for
  • timeout_seconds (integer): Maximum wait time
  • poll_interval_seconds (integer): Check interval

​
Fail

Explicitly fail the pipeline with a message. Parameters:
  • message (string, required): Failure message
  • error_code (string): Error code for categorization

​
Deployment Actions

​
Deploy Model

Create a model deployment endpoint. Parameters:
  • model_id (string, required): Model to deploy
  • cluster_id (string, required): Target cluster
  • deployment_name (string, required): Deployment identifier
  • replicas (integer): Initial replica count
  • auto_scaling (object): Autoscaling configuration
Outputs:
  • deployment_id: Created deployment ID
  • endpoint_url: Inference endpoint URL
  • status: Deployment status
Example: 
{
    "action": "deployment_create",
    "params": {
        "model_id": "{{steps.add_model.output.model_id}}",
        "cluster_id": "cluster_prod",
        "deployment_name": "llama-inference",
        "replicas": 2
    }
}

​
Scale Deployment

Adjust the number of replicas for a deployment. Parameters:
  • deployment_id (string, required): Deployment to scale
  • replicas (integer, required): Target replica count
Outputs:
  • previous_replicas: Replica count before scaling
  • current_replicas: New replica count

​
Configure Rate Limiting

Set rate limits for a deployment endpoint. Parameters:
  • deployment_id (string, required): Target deployment
  • rate_limit (integer, required): Requests per time window
  • time_window (string): Time window - second, minute, hour
  • algorithm (string): Algorithm - token_bucket, fixed_window

​
Delete Deployment

Remove a deployment and release resources. Parameters:
  • deployment_id (string, required): Deployment to delete

​
Model Operations Actions

​
Add Model

Add a model from HuggingFace or URL to the registry. Parameters:
  • model_uri (string, required): Model identifier or URL
  • model_name (string, required): Display name
  • model_source (string, required): Source - hugging_face, url, local
  • max_wait_seconds (integer): Maximum download wait time
Outputs:
  • model_id: Created model ID
  • model_name: Model name
  • size_bytes: Model size
Example: 
{
    "action": "model_add",
    "params": {
        "model_uri": "meta-llama/Llama-3.2-1B-Instruct",
        "model_name": "Llama-3.2-1B-Instruct",
        "model_source": "hugging_face"
    }
}

​
Add Cloud Model

Add a cloud-hosted model (OpenAI, Anthropic, etc.). Parameters:
  • provider (string, required): Cloud provider
  • model_id (string, required): Provider’s model ID
  • api_key (string): API key (use secrets)

​
Model Benchmark

Run performance benchmarks on a model. Parameters:
  • model_id (string, required): Model to benchmark
  • test_prompts (array): List of test prompts
  • metrics (array): Metrics to measure
Outputs:
  • latency_p50: 50th percentile latency
  • latency_p99: 99th percentile latency
  • throughput: Requests per second

​
Delete Model

Remove a model from the registry. Parameters:
  • model_id (string, required): Model to delete

​
Cluster Actions

​
Cluster Health Check

Verify cluster status and readiness. Parameters:
  • cluster_id (string, required): Cluster to check
  • timeout_seconds (integer): Health check timeout
Outputs:
  • status: Cluster status - healthy, degraded, unhealthy
  • node_count: Number of active nodes
  • available_resources: Available CPU/GPU/memory
Example: 
{
    "action": "cluster_health",
    "params": {
        "cluster_id": "cluster_prod",
        "timeout_seconds": 30
    }
}

​
Integration Actions

​
HTTP Request

Make external API calls. Parameters:
  • url (string, required): Request URL
  • method (string): HTTP method - GET, POST, PUT, DELETE
  • headers (object): Request headers
  • body (object): Request body
  • timeout_seconds (integer): Request timeout
Outputs:
  • status_code: HTTP status code
  • response_body: Response data
  • headers: Response headers
Example: 
{
    "action": "http_request",
    "params": {
        "url": "https://api.example.com/webhook",
        "method": "POST",
        "headers": {
            "Content-Type": "application/json"
        },
        "body": {
            "event": "deployment_complete",
            "deployment_id": "{{steps.deploy.output.deployment_id}}"
        }
    }
}

​
Send Notification

Send alerts via email, Slack, or Teams. Parameters:
  • message (string, required): Notification message
  • channel (string, required): Channel type - email, slack, teams
  • recipients (array): List of recipients
  • priority (string): Priority level
Example: 
{
    "action": "notification",
    "params": {
        "message": "Deployment {{steps.deploy.output.deployment_name}} completed successfully",
        "channel": "slack",
        "priority": "normal"
    }
}

​
Common Patterns

​
Referencing Step Outputs

Access outputs from previous steps: 
{{steps.step_id.output.field_name}}

​
Conditional Execution

All actions support conditional execution: 
{
    "action": "notification",
    "condition": "steps.deploy.output.status == 'active'",
    "params": {...}
}

​
Error Handling

Configure retry behavior: 
{
    "action": "deployment_create",
    "retry": {
        "max_attempts": 3,
        "backoff_multiplier": 2
    },
    "params": {...}
}

​
Next Steps

Pipeline Concepts

Learn about DAGs and workflows

Quick Start

Build your first pipeline