Create a generic agent profile - Fluid Topics - Latest

Use Agentic chatbot profiles for the Generic agentic retrieval-augmented generation (RAG) web service.

To create an Generic agent AI profile:

1. Select New profile, and select Generic agent.

2. Enter or select:

   Field	Description
   Profile name	An AI profile name.
   Profile ID	A unique profile ID. The profile name and ID must be different from existing profiles.
   Provider	Choose one of Amazon Bedrock, Anthropic, Azure OpenAI, or OpenAI.
   Model	An LLM model from the selected provider. For Azure OpenAI, provide a host and model name. For Amazon Bedrock, provide a model name only. It is mandatory to select an agentic-ready model. For example, gpt-5 and up for OpenAI. Deployment of an agentic-ready model for Azure is only available worldwide yet. Consult the list of supported regions to make sure it is possible to use an agentic ready model in your region. The LLM model dropdown menu is not exhaustive. It is possible to enter any value, as long as it matches a valid model from the selected LLM provider. For example, for a refined model only known by the person configuring the profile. Use the Test connection button to check the profile configuration.
   Authentication	An API key from the provider, or OAuth 2.0 client credentials.
   Agent instructions	Instructions for the agent. The instructions define specific rules and the conversation scope for the bot. For more information, see the Advanced settings for Generic agent profiles topic. For optimal results, add specific instructions that are relevant to your context. Below are a few examples: - If the user is asking about product X, search for release notes X before answering. - Make sure you have properly identified the product the user is referring to. - Suggest to create a support ticket and provide a link to support.example.com if the user found a critical bug. - If the relevant documentation contains a topic named "Prerequisites" or "Before you begin", provide its content to the user.

3. Enter the JSON schema that the agent's responses must conform to.

   For example, the following codeblock orders the agent to give users a list of products the user may be talking about, with a confidence score:

   {
       "type": "array",
       "items": {
           "type": "object",
           "properties": {
               "metadata_value": {
                   "type": "string"
               },
               "confidence": {
                   "type": "number",
                   "minimum": 0,
                   "maximum": 1
               }
           },
           "required": [
               "product_name",
               "confidence"
           ],
           "additionalProperties": false
       }
   }
   

   The agent would answer like in the following example:

   [
       {
           "product_name": "iPhone 15 Pro",
           "confidence": 0.92
       },
       {
           "product_name": "iPhone 15",
           "confidence": 0.76
       }
   ]
   

4. Optional: Select tools for the agent to use.

5. Optional: Add inferred filters.

   Inferred filters allow the agent to use metadata filters for their queries. Administrators must enter a description for how and when the agent should use the metadata filter.

   If the agent runs a query with filters and gets no response, it runs the query again.

6. Optional: Add cluster priority.

   Cluster priority uses metadata to indicate which variant of a document is the most relevant in a document cluster. For example: "version": "latest".

   • If the variant does not exist, the agent ignores the rule.
   • Rules are applied in order.

7. Optional: Enter advanced settings.

8. Select Save.

Once created, ADMIN users can copy the generic agent profile ID to use in AI features.