Workflows are the core of AI agents. A workflow represents a collaborative group of agents working together to achieve a set of tasks. Each workflow defines the strategy for task execution and agent collaboration.
Each agent can be standalone, or can have a arsenal of tools. From CrewAI's docs, an agent is defined as an autonomous unit that can:
- Perform specific tasks
- Make decisions based on its role and goal
- Use tools to accomplish objectives
- Communicate and collaborate with other agents
- Maintain memory of interactions
- Delegate tasks when allowed
You can configure an agent with the following parameters:
- Role: Defines the agent’s function and expertise within the workflow.
- Goal: The individual objective that guides the agent’s decision-making.
- Backstory: Provides context and personality to the agent, enriching interactions.
- Tools: The tools that the agent can use to accomplish its goal.
A task is a unit of work that an agent can perform. Tasks provide all necessary details for execution, such as a description, the agent responsible, required tools, and more, facilitating a wide range of action complexities.
Tasks can be executed in two ways:
- Sequential: Tasks are executed in the order they are defined
- Hierarchical: Tasks are assigned to agents based on their roles and expertise
A task can be configured with the following parameters:
- Description: A clear, concise statement of what the task entails.
- Expected Output: A detailed description of what the task’s completion looks like.
- You would be asked to pick a workflow name. For the sake of this example, we'll assume that you're creating a fresh workflow and not from a template.
- You're presented with two switches:
- Is Conversational: If enabled, the workflow will be a conversational workflow.
- Manager Agent: If enabled, the workflow will have a manager agent. Having a manager agent allows the workflow to delegate tasks to suitable agents. Enabling this would change the workflow type from
sequentialtohierarchical.
- You can add agents to the workflow. For each agent, you can create a tool from existing tool template, or just a brand new tool.
- You can add tasks to the workflow. For each task, you will need to select the agent that should execute the task, if a manager agent is not selcted in the previous step.
- In the next step, you'll asked to conifgure the tools created in the previous step.
- Once tools are configured, you can test the workflow in-studio by providing inputs as defined in your tasks.
- The tool configuration parameters(which might contain sensitive information like API keys) are stored securely in your browser's local storage.
- Once you're happy with the workflow, you can deploy it as a separate application in the workbench.
- While deploying you can save the workflow as a template for future use.
NOTE: At any point in time during the above steps, you can drop from the workflow editor, and the workflow would be saved savely as a draft.
A deployed workflow is a stanadlone application that no longer allows you to edit the workflow. It is meant to be a production-ready application, ready to be used by end-users(or other stakeholders within your organization).
Each deployed consists of 2 components:
- The workflow backend is served my a AI Workbench model.
- The workflow frontend is served by a AI Workbench application.
A workflow when deployed, would give you deeplinks to access the application, the workbench configuration of the model & workbench configuration of the application.
Workflow templates are pre-configured workflows that can be used as a starting point for new workflows. They are useful for quickly creating new workflows with a predefined structure.