> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mindset.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Workflows

> Workflows in the AMS

**NOTE:** Available to AMS customers only

## Structured, Multi-Step Processes for Your Agents

Agent workflows let you create and manage multi-step agent processes that take full advantage of all Mindset AI’s capabilities.\
You can make these very deterministic or give the agent scope to make decisions as part of these workflows.

**Our goal**: Make it incredibly easy and reliable for a human to say *"Help me with X"* and have the agent figure out how to get it done.

Workflows you build let your agents carry out structured, multi-step processes — like following a recipe from start to finish. This makes complex tasks **faster, more reliable, and more consistent** for your users.

## Core Workflow Advantages

* **Vibe code agentic workflows** – Easily chain together multiple steps and tasks to create complex agent workflows, all by speaking to AI.
* **Context management** – Manage and share information between tasks using variables that persist throughout the workflow execution.
* **Sequential, single flow architecture** – Built on a sequential single flow model (vs multi-agent systems), as it is best for the majority of tasks, more reliable, and easier to scale in production.
  * Many complex multi-agent architectures (like those in some n8n or LinkedIn automation setups) are unnecessarily complicated and hard to scale in SaaS environments.

## Key Features

### Create

A workflow is a **pre-built sequence of steps** your agent follows from start to finish.\
Steps can include:

* Asking the user questions
* Retrieving information
* Generating responses
* Showing messages

You can control what information is remembered or captured during a step via **variables**.

### Step Types: Your Building Blocks

#### 1. Fulfill goal – Get info from the user

* Questions, reflections, answers, input
* Perfect for collecting requirements, asking clarifying questions, or getting user approval\
  **Example**: `"What industry is your company in?"` or `"Please describe your main challenge with customer retention."`

#### 2. RAG search – Retrieve relevant facts

* Pull relevant information from ingested content
* Retrieve documentation, find similar cases, or access company policies\
  **Example**: Searching for industry-specific case studies after learning the user's industry.

#### 3. Call LLM – Let the LLM do the work

* Analyze, summarize, or give feedback based on information from the user or RAG search
* Generate recommendations, analyze data, or create custom content\
  **Example**: Analyzing gathered requirements and generating a customized proposal.

#### 4. Print step – Communicate with the user

* Tell the agent exactly what to say: instructions, congratulations, transitions between stages
* Complete control over user communication\
  **Example**: `"Based on your requirements, here's your personalized implementation plan..."`

### Variables System: Store and Reuse Values

Store and reuse values between steps using the `@variable_name` format.\
**Example**:

1. Collect the user's name in a fulfill goal step → `@user_name`
2. Use `"Hello @user_name"` in a later print step.

## Preview

You can test workflows in the preview section before assigning them to an agent.\
Each workflow gets a **unique ID**, making it easy to manage.

## Deploy

Any workflow you build should be **added to an MCP server** — think of these as a GDrive or toolbox for agents to use (e.g., `Sales actions`, `Coaching actions`, `Content actions`).

* You can assign **up to 25 workflows** to a single agent via MCPs.

  [Read more about MCP servers](/amp/MCPservers)

## Getting Started

### Prerequisites

* Must be using SDK 2.0 or later.

### Current Limitations

* **Maximum steps per workflow**: 50
* **Editing steps**: You can’t edit a single step in isolation yet; changes must be made through the Workflow Builder or by regenerating the workflow with the agent’s help.