Automating Multi-Step Agent Workflows

Design and run durable multi-step pipelines where each step invokes an AI agent, with async execution, Human-in-the-Loop pausing, and workflow cloning.

Workflows let you chain agent actions into repeatable, multi-step pipelines. Each step in a workflow triggers an agent run, passing input forward through the sequence. Workflows execute asynchronously — you receive a job ID immediately and poll for progress — making them suitable for long-running automation tasks like content pipelines, research flows, and multi-stage data processing.

What is a workflow?

A workflow is a named, ordered sequence of steps. Each step references an agent ID and optional configuration. When you execute a workflow, Agent Manager runs each step in order, feeding the output of one step as the input to the next. The state of every run is persisted to the database, so workflows survive restarts and can be resumed after a human approval pause.

Content pipelines

Research a topic with web_agent, summarize with procurator_assistant, then draft a post.

Research flows

Gather data across multiple sources, analyze findings, and produce a structured report.

Data processing

Extract, transform, and validate data across agents with different tool sets.

Creating a workflow

POST /api/v1/workflows

Request body (WorkflowDTO):

name · string · required

A human-readable name for the workflow.

description · string

Optional description of what the workflow does.

Example:

{
  "name": "Tech News Briefing",
  "description": "Searches for tech news, summarizes it, and drafts a newsletter section."
}

Response — the created workflow with its generated ID:

{
  "id": "wf-3a7bc9",
  "name": "Tech News Briefing",
  "description": "Searches for tech news, summarizes it, and drafts a newsletter section."
}

Adding steps

After creating a workflow, add steps to define what each agent should do. Steps execute in the order they are added (by stepOrder).

POST /api/v1/workflows/{id}/steps

Request body (WorkflowStepDTO):

agentId · string · required

The ID of the agent to invoke for this step.

action · string · required

The prompt or instruction sent to the agent for this step.

stepOrder · number

The position of this step in the sequence, starting from 1. Steps execute in ascending order.

Example — three-step content pipeline:

1. Step 1: Search

   {
     "agentId": "web_agent",
     "action": "Search for the latest news about artificial intelligence in healthcare",
     "stepOrder": 1
   }
   
2. Step 2: Summarize

   {
     "agentId": "procurator_assistant",
     "action": "Summarize the search results into 3 bullet points",
     "stepOrder": 2
   }
   
3. Step 3: Draft

   {
     "agentId": "procurator_assistant",
     "action": "Turn these bullet points into a 2-paragraph newsletter section",
     "stepOrder": 3
   }
   

Executing a workflow

Trigger an asynchronous run of the workflow:

POST /api/v1/workflows/{id}/run

Request body:

input · string · required

The initial input passed to the first step.

sessionId · string

Optional session ID for context continuity. A new session is created if omitted.

Example:

{
  "input": "artificial intelligence in healthcare 2025",
  "sessionId": "session-news-001"
}

Response — returned immediately with 202 Accepted:

{
  "jobId": "job-f9a44e",
  "workflowId": "wf-3a7bc9",
  "sessionId": "session-news-001"
}

The workflow runs asynchronously. The jobId in the response is an internal queue job identifier. Use the workflow runs endpoint (below) to track execution progress by workflowId.

Checking run status

List all historical runs for a workflow, newest first:

GET /api/v1/workflows/{workflowId}/runs

Supports pagination via ?page=0&size=20.

Run response fields:

id · string

Unique identifier for this workflow run.

workflowId · string

ID of the parent workflow.

sessionId · string

The session used for this run.

status · string

Current run status: RUNNING, COMPLETED, FAILED, or PAUSED.

currentStepOrder · number

The step number currently being executed (or last executed if the run is paused or complete).

durationMs · number

Total elapsed time in milliseconds (computed from createdAt to updatedAt).

createdAt · string

ISO 8601 timestamp when the run was created.

updatedAt · string

ISO 8601 timestamp of the last status change.

Example response:

{
  "content": [
    {
      "id": "wr-55cd12",
      "workflowId": "wf-3a7bc9",
      "sessionId": "session-news-001",
      "status": "COMPLETED",
      "currentStepOrder": 3,
      "durationMs": 14320,
      "createdAt": "2026-05-06T10:00:00Z",
      "updatedAt": "2026-05-06T10:00:14Z"
    }
  ],
  "totalElements": 1,
  "totalPages": 1
}

Resuming a paused workflow

If a step involves an agent tool that requires human approval, the workflow pauses at that step and the run enters PAUSED status. Resume it with the approved output:

POST /api/v1/workflows/runs/{runId}/resume

Request body:

output · string · required

The human-approved content to inject as the output of the paused step. This value is passed as input to the next step in the sequence.

Example:

{
  "output": "Approved summary: AI is transforming diagnostics and drug discovery."
}

Response — returned immediately with 202 Accepted:

{
  "jobId": "job-cc8812",
  "runId": "wr-55cd12"
}

A paused workflow run does not time out automatically. Ensure your application monitors for PAUSED runs and provides a mechanism for operators to review and resume them.

Cloning a workflow

Create an independent copy of an existing workflow with all its steps:

POST /api/v1/workflows/{id}/clone

No request body is required. The cloned workflow is created with the suffix (Copy) appended to its name and a new unique ID.

Example response:

{
  "id": "wf-7d1ee2",
  "name": "Tech News Briefing (Copy)",
  "description": "Searches for tech news, summarizes it, and drafts a newsletter section."
}

Use cloning to create environment-specific variants of a workflow (e.g., staging vs. production) without rebuilding steps from scratch.

Workflow management endpoints

Method	Endpoint	Description
GET	/api/v1/workflows	List all workflows (paginated).
GET	/api/v1/workflows/{id}	Get workflow by ID.
POST	/api/v1/workflows	Create a new workflow.
PATCH	/api/v1/workflows/{id}	Update workflow name or description.
DELETE	/api/v1/workflows/{id}	Delete a workflow and its steps.
GET	/api/v1/workflows/{id}/steps	List steps for a workflow.
POST	/api/v1/workflows/{id}/steps	Add a step to a workflow.
DELETE	/api/v1/workflows/{id}/steps/{stepId}	Remove a step.
POST	/api/v1/workflows/{id}/clone	Clone a workflow with all its steps.
POST	/api/v1/workflows/{id}/run	Execute a workflow asynchronously.
GET	/api/v1/workflows/{workflowId}/runs	List run history for a workflow.
POST	/api/v1/workflows/runs/{runId}/resume	Resume a paused workflow run.