API Agent overview
An overview of the API Agent user interface and its key components.
Connection details
After you log in, you can view your connection details at the upper-left bar. You can click on the text to the left of AI to access more specific information, which includes the API Connect management URL and your provider organization.
New chat
- Click the IBM API Agent menu and select New chat
to start a new session.
- Select a provider organization. You must already be a member of this organization, and the API Agent must be enabled
there for you. This allows you to use the API Agent.
Click Continue. A welcome message is displayed.
-
In the Type something field, enter text to start interacting with API Agent.
Plan
- Plans that do not modify the system (read-only operations) are executed automatically and do not require user approval.
- Plans that modify the system state (such as creating, updating, or deleting resources) requires explicit user approval before execution.
The following example shows the plan for user message of create test cases for the petstore api and run the test cases.
Viewing and editing the plan
After a plan is displayed, you can view and modify the suggestions for each tool. If the default suggestions that are provided by the API Agent are satisfactory, you can approve them and continue.
To make any changes, select Edit plan to open the Edit plan window, where you can make changes to the values.
Approval
User approval is required to execute any tool that makes changes to the system state. These tools cannot be run without explicit user consent.
Understanding responses
Each response is a result that is displayed for the execution of a single tool. It includes a summary, expand the arrow option to view a detailed system response.
After you click Approve, the tool runs, and displays a response. The following is an example of a successful run.
Errors
If your prompt requests an action that the API Agent cannot perform because the system lacks the appropriate set of tools for the request. The API Agent states that the request cannot be completed with the current tools. It will then list the available tools and their brief descriptions. Any other errors are execution errors of the tools. In such cases, the API Agent provides a summary of the error that was encountered during the tool's run time.
Aborting
You have the option to disregard the generated plan and initiate another request. If you choose to do so, the previous prompt and action are disregarded.
Commands
You can run some of the available tools directly by using
/commands. If you start typing forward slash (/), you can view
the list of available commands.
To pass parameters, you have to give values in key:value format. Space is
allowed before value, which makes it easier to specify file to upload. The following is an example
of using /command.
Uploading and downloading files
Some commands need files to be uploaded. The file must be in your Visual Studio Code workspace for uploading.
Entering @ in the Type something field displays a list of available files. As you input characters, the list is filtered accordingly. You can select a file at any point during this process.
Where the @file name is listed, you can upload a file from
the local workspace.
Whenever api_name is mentioned, you can use it with id,
name:version, or the name of a draft API. Using just the name is
acceptable as long as there is only one draft API with that specific name. However, using the title
of an API is not permissible. The same guidelines apply when product_name is
mentioned.
Some commands allow you to download files from the response, which is indicated by a download icon.
