Building a custom extension

The Custom tab on the CloudRadial AutomationAI Extensions page lets you build an extension that is private to your own workspace. You can describe a service and let AI research and build it, or author the definition by hand and upload it. This article is for Admins and Owners.

• The Custom Tab
• Generating an Extension With AI
• Authoring or Importing a Definition by Hand
• Reviewing, Downloading, and Deleting

The Custom Tab

Open Extensions and select the Custom tab. It lists your tenant-private extensions, each shown with the Private tier, its version, a description, and its skill and tool counts. A custom extension is active for your runners with no separate install step. Select AI Generate to add one.

Generating an Extension With AI

The AI Generate Extension panel's primary path is to describe what you want and let AI build it.

1. Enter the app or service name in App or service name.
2. Optionally paste one or more links to its API documentation in Documentation link(s), one per line. If you do not provide a link, AI searches for the documentation.
3. In What should the agent be able to do?, describe the actions you want in plain language.
4. Select AI Generate. The button reads Generating… while AI researches and builds the extension, with a progress line below it.

AI reads the documentation and works out the connection details, then saves a private extension for you to review. If it needs more detail, it asks one plain-language question in the panel; add the detail and select AI Generate again.

Authoring or Importing a Definition by Hand

To write the definition yourself, expand Author manually or import a .yml in the same panel. From there you can:

• Select Download template to get a starter YAML file
• Select Copy AI editor prompt or Download .md to get a prompt that documents the schema for your own AI editor
• Select Upload .yml to upload your edited file

An uploaded file is validated and saved by slug. If a custom extension with the same slug already exists, or the file produces warnings, AutomationAI asks you to review and confirm before it replaces the definition. If the file is rejected, the panel shows the reason, so you can fix the definition and upload it again.

OAuth Client Credentials Sign-In Options

Some services need more than a client ID and client secret when an extension signs in with OAuth client credentials. Two optional settings in the definition's sign-in (auth) section cover this:

• tokenParams adds extra fields to the token request, such as an audience the service requires. For example: "tokenParams": {"audience": "https://api.example.com"}.
• "tokenBodyFormat" sets how the token request is sent: "form" (the default) or "json", for services that accept the token request only as JSON

A few rules apply:

• Each tokenParams value is sent exactly as written. Secret names are not looked up there, so never put a secret or credential in it.
• Every value must be text
• The fields the runner sends itself are reserved and cannot be set in tokenParams, in any capitalization: grant_type, client_id, client_secret, scope, username, and password
• A definition that breaks one of these rules, or sets tokenBodyFormat to anything other than form or json, is rejected, and the panel shows why

Both settings require a runner on version 2.5.0 or later. An earlier runner ignores them, so the service usually refuses the token request. A runner on version 2.0 or later updates itself automatically unless it is pinned, set to manual updates, or outside its update window; in those cases, check the runner's version on the Runners page and use Update now. See Upgrading a runner to a new version.

Reviewing, Downloading, and Deleting

After a successful generate or import, the new extension's detail panel opens so you can review its skills and tools. On each row in the Custom list you can:

• Select the row to open its detail panel
• Select Download to save its definition as YAML for local editing and re-upload
• Select Delete to remove it and its skills and tools, after confirming in the dialog