Skip to main content

​
Overview

Workflows enable you to define complex, multi-step automated processes that orchestrate agents, tools, and external services. Built on Temporal, workflows provide reliable, durable execution with automatic retries and error handling.

​
Workflows

Workflows are declarative definitions of multi-step processes that can:
  • Execute multiple tasks in sequence or parallel
  • Coordinate multiple agents
  • Handle errors and retries automatically
  • Maintain state across long-running operations
  • Integrate with external systems
Key features:
  • Durability: Workflows survive crashes and restarts
  • Observability: Track execution state and history
  • Scheduling: Run workflows on a schedule or trigger basis
  • Versioning: Safely update workflow definitions

​
Runners

Runners provide the execution infrastructure for workflows and agents. They:
  • Execute workflow steps in isolated environments
  • Provide compute resources for agent execution
  • Handle scaling and resource management
  • Support different runtime environments (containers, VMs, etc.)
Runner types:
  • Kubernetes runners: Execute in Kubernetes pods
  • Container runners: Execute in Docker containers
  • Cloud runners: Execute in cloud VMs (AWS, GCP, Azure)

​
Common Workflow Patterns

​
Sequential Workflow

name: "Deploy Application"
steps:
  - name: "Run Tests"
    agent: "test-runner"
  - name: "Build Docker Image"
    agent: "builder"
  - name: "Deploy to Production"
    agent: "deployer"
    requires: ["Run Tests", "Build Docker Image"]

​
Parallel Workflow

name: "Multi-Cloud Deployment"
parallel:
  - name: "Deploy to AWS"
    agent: "aws-deployer"
  - name: "Deploy to GCP"
    agent: "gcp-deployer"
  - name: "Deploy to Azure"
    agent: "azure-deployer"

​
Event-Driven Workflow

name: "Incident Response"
trigger:
  type: "webhook"
  source: "alertmanager"
steps:
  - name: "Analyze Alert"
    agent: "analyzer"
  - name: "Take Action"
    agent: "remediator"
  - name: "Notify Team"
    agent: "notifier"

​
Workflow Management

​
Create a Workflow

POST /api/v1/workflows
{
  "name": "Deploy Application",
  "description": "Multi-step deployment workflow",
  "steps": [
    {
      "name": "run-tests",
      "type": "agent",
      "agent_id": "test-agent-uuid"
    },
    {
      "name": "build-image",
      "type": "agent",
      "agent_id": "builder-agent-uuid"
    },
    {
      "name": "deploy",
      "type": "agent",
      "agent_id": "deployer-agent-uuid"
    }
  ],
  "team_id": "team-uuid"
}
Required fields: name Note: Steps are executed sequentially by default

​
Start Workflow Execution

POST /api/v1/workflows/{workflow_id}/start
{
  "input": {
    "environment": "production",
    "version": "v1.2.3"
  }
}

​
Best Practices

  1. Idempotency: Design workflow steps to be safely retried
  2. Error Handling: Define explicit error handling and compensation logic
  3. Timeouts: Set appropriate timeouts for each step
  4. Monitoring: Use workflow history for debugging and auditing
  5. Versioning: Version workflows to support safe updates

​
Next Steps

Explore the workflow and runner endpoints to learn about creating, executing, and monitoring workflows.