Get Started with Agents

Note For access, please talk to our Support team.

Agents in Agent OS are intelligent tools designed to execute complex, multi-step workflows that go beyond traditional, linear automation. They leverage AI models and connectors to perform dynamic tasks, reasoning, and decision-making.

This guide provides a hands-on introduction by walking you through the creation of an AI-powered agent. This agent automatically discovers, analyzes, and publishes breaking news, demonstrating an end-to-end automation cycle.

The agent you build performs the following complex workflow:

1. Trigger: Activates the process via a simple HTTP URL call.
2. Discovery and analysis: Searches for the top three breaking AI news items, generates a summary of each, and analyzes its potential impact on the IT industry.
3. Content automation: Creates a separate content entry in your Contentstack CMS for each news item.
4. Notification: Sends a Slack notification with the compiled summaries and impact analysis to a designated channel.

1. Contentstack account
2. Agent plan for your organization
3. Admin/Owner access for the Contentstack stack

Let's start by logging into the Contentstack account and following the steps given below:

1. After logging in, click the App Switcher icon, then select Agent OS from the list.
   Click to enlarge

Let’s see the two ways to create an agent:

1. Create an agent in a new project.
2. Create an agent in an existing project.

1. On the Agent OS projects page, click + New Project.
2. Enter a Name and an optional Description, then click + Create Project.
3. From the Agent OS Dashboard page, click + New Agent.
4. In the Create Agent modal, click Skip, I'll create manually. Enter a suitable Title and a Description for your agent. Click the Create Agent button.
   Click to enlarge

   Note You can also create an agent using the Automated Setup, where you provide a description and the system automatically configures the trigger, tools, and instructions.

5. You are redirected to the Agent Builder page, where you can add the Trigger, Tools, and Instructions.
   Click to enlarge

1. Navigate to your project.
2. From the Dashboard page, click the Create drop-down.
   Click to enlarge
3. From the dropdown, select + New Agent button.
   Click to enlarge
4. In the Create Agent modal, click Skip, I'll create manually. Enter a suitable Title and a Description for your agent. Click the Create Agent button.
   Click to enlarge

   Note You can also create an agent using the Automated Setup, where you provide a description and the system automatically configures the trigger, tools, and instructions.

5. You are redirected to the Agent Builder page, where you can add the Trigger, Tools, and Instructions.
   Click to enlarge

Let’s see how to configure the Trigger, Tools, and Instructions for the agent.

The Agent Builder comprises three sections: Trigger, Tools, and Instructions.

Trigger defines the events that start your agent’s workflow, allowing it to run automatically.

1. In the Triggers section, click + to add the trigger. A side panel opens.

   Additional Resource Refer to the Triggers documentation to learn more.

   Click to enlarge
2. Select the HTTP Trigger and click Save. Once done, you are ready to use the trigger.

   Additional Resource Refer to the HTTP Trigger documentation to learn more.

3. Click the vertical ellipsis to edit the configuration or replace the trigger. Selecting either option opens a side panel where you can modify the existing trigger configuration or replace it with a new trigger.
   Click to enlarge

The Tools section lets you define what your agent can do by selecting the tools, automations, and agents it can use to perform tasks.

For our use case, we will select the three tools: Create an Entry action, Slack connector, and ChatGPT: Web Search.

Let’s see how to add all three tools:

1. In the Tools section, click + Add to add the tools.
   Click to enlarge

   ChatGPT: Web Search

   1. A side panel opens. Under the Tools category, select ChatGPT: Web Search. Authenticate your ChatGPT account, then select a Model. For the remaining fields, let AI select the values. Click Save.

      When you select a tool:

      1. A modal opens with two configuration options:

         1. Let AI select data: The agent automatically extracts the values for the action fields from the prompt.

            For example, Agent Prompt: Retrieve values from the Sample Stack and choose the AI Intelligence entry.

            These values are selected dynamically at runtime.

         2. Add custom data: You can manually select predefined values using the Lookup drop-down.

            Note All actions are shown in unselected mode by default, and you can choose any available connector as needed.

            Click to enlarge

      Additional Resource For more information, refer to the ChatGPT connector documentation.

   CMS: Create an Entry action

   1. Click + Add to add the Create an Entry action.
   2. In the side panel, under the Tools category, select CMS.
   3. In the Entry category, select the Create an Entry action.
   4. On the Create an Entry configuration screen, authenticate your Contentstack account.
   5. From the Select Stack drop-down, select Add custom data, then select a stack from the Lookup drop-down.
   6. For the remaining fields, let AI select the values. Once complete, click Save.
      Click to enlarge

   Slack Connector:

   1. Click + Add to add the Slack connector.
      Click to enlarge
   2. Select the Send a Message action.
   3. Add your Slack account.
   4. From the Channel drop-down, select Add custom data, then choose a channel from the Lookup drop-down.
   5. For the rest of the fields, let AI select the values. Click the Save button.
      Click to enlarge
2. Once done, you see all the added tools.
   Click to enlarge

   Additional Resource Refer to the Connectors documentation to learn more.

Instructions, the most important component, define the agent’s role and rules. It controls how the agent behaves and what “good output” looks like.

1. Navigate to the Instructions section.
2. Add the instructions required for agent execution, as shown below:
3. Use / to add tools to your set of instructions.
   Click to enlarge

Once the agent configurations are set, click the Save button from the top-right navigation to save the agent.

Click to enlarge

To publish the agent, follow the steps below:

1. Click the Publish button from the top-right navigation panel.
   Click to enlarge
2. In the Publish Agent pop-up, click the Publish button.
   Click to enlarge

Once the agent is published, hit the HTTP URL. You see three entries created in the CMS.

To edit the agent, follow the steps below:

1. To edit the agent configurations, click the Edit button.
   Click to enlarge
2. Once you click the Edit button, the agent is available in the Draft mode to edit.

Unpublishing an agent revokes the latest changes. To unpublish the agent, follow the steps below:

1. Click the Unpublish button from the top-right navigation panel.
   Click to enlarge
2. In the Unpublish Agent pop-up, click the Unpublish button.
   Click to enlarge

Note You can view the agent versions by clicking the Version icon.

Let’s see the steps to create an agent using a predefined template:

1. In the Create Agent modal, click the Browse Template button.
   Click to enlarge
2. In the Browse Templates pop-up, select a template to get started. You can hover over any template and click the View template link to view its configuration.
   Click to enlarge
3. Click Use Template to use an existing template. You are redirected to the Agent Builder screen.
   Click to enlarge
4. You are redirected to the Agent Builder page, where you can see the Trigger, Tools, and Instructions configured for your template.
   Click to enlarge

The Executions view provides a complete and transparent breakdown of how an automation or agent run performed from start to finish, helping users understand outcomes, performance, and resource usage.

Includes:

• Status: Displays a clear execution status (for example, Success).
• Started At: Indicates the start time of the execution.
• Total Duration: Displays the total duration of the run.
• Tools Used: External tools or capabilities the agent invoked during an execution.

On clicking an individual execution

Execution steps
The Execution Steps timeline shows every action performed during the run in chronological order, making it easy to trace agent behavior and identify performance bottlenecks.

Includes:

• Step-by-step execution flow (for example, web search, content creation, message delivery)
• Metrics:
  • Started At: Shows the exact date and time when the execution began.
  • Duration: Indicates how long the execution took to complete from start to finish.
  • Total Tokens: Represents the total number of tokens consumed during the run, helping track usage and cost.
  • Model: Identifies the AI model used to execute the task.
• Input and Output:
  • Input:
    • Displays the exact prompt or instructions provided to the agent.
    • Shown in JSON format for reproducibility.
  • Output:
    • Displays structured execution results in JSON format.

• Success but no result: Check output data.
• Link check errors (405/403): Ensure fallback from HEAD to GET.
• Intermittent failures: Compare timestamps with rate limits or API issues.

The General section lets you define your agent’s identity with a title and description, so its purpose is clear from the start.

1. You see a Title (required) and an optional Description for the agent’s purpose.

   The app automatically picks a name and description for your agent based on what you entered in the Create Agent step. If you want, you can edit the name or update the description.

2. Click the Save button.
   Click to enlarge

The AI Model screen is where you choose and connect the underlying model that powers your agent, ensuring it can reason and respond effectively. You see Contentstack Managed (Organization Default).

Click to enlarge

By clicking the Change link, you see two configurations: Organization Default and Custom Configuration.

Custom Configuration:

1. In the Select Model Provider (required) drop-down , select the LLM provider for your agent. Currently supported providers include:

   • Azure OpenAI
   • Gemini
   • OpenAI
   • Google Vertex
   • Anthropic

   For our use case, we used OpenAI.

2. In the Agent Authentication field, add or select an account to authenticate the agent. To create a new key, click + Add API Key and sign in with your model provider’s API Key and Organization ID (for OpenAI). If you select Gemini, you see Google Vertex authentication options; if you select OpenAI, you see ChatGPT authentication options.
3. In the Select AI Model field, select which underlying AI model the agent should use to run tasks or generate responses.
4. Click the Save button to save the configuration.
   Click to enlarge

To delete an agent, click the Delete Agent button. A pop-up appears. Enter DELETE in the input field and click the Delete Agent button.