Configure the management of your agents

The Agents Management page defines how SmartHub executes agentic Retrieval-Augmented Generation (RAG) during conversational search. It allows administrators to configure the individual agents that control reasoning, retrieval, response generation, and insight extraction across the search workflow.

Rather than treating conversational search as a single LLM call, SmartHub uses an agent-based architecture. Each section on this page represents a distinct responsibility in that architecture, giving you fine-grained control over how queries are interpreted, processed, and answered. Once configured, you can enable the relevant agents that you need to process your user's queries.

How Agents Management Fits into Conversational Search

When a user submits a conversational search query, SmartHub processes it through a multi-stage agent workflow, allowing conversational search to process the query in the following ways:

1. The Orchestrator agent determines how AI is used for a user's search query, and whether the indexed content will be used to retrieve an answer. For example, if the query is generic, such as "What happened on this day in history?", the AI will not use the indexed content. Conversely, if the query relies on indexed content, such as "What Jira tickets were created after <date>?", the AI will search the index to retrieve the answer. This agent is required to retrieve an answer

2. The Query Processor agent determines the intent of the users search query before retrieval, cleans the user search query, and establishes additional details related to the search query to regulate the response to specific sources, filters, and user instructions.

3. The Response Generator agent analyzes the information collected from the Orchestrator and Query Processor agents to generate the response. This agent is required to retrieve an answer.

4. The Result Insight agent determines additional suggestions that the AI can make based on the response generated, such as filters or follow-up questions.

Configure an agent group

You can configure an agent group to customize your Orchestrator, Query Processor, Response Generator, and Result Insight agents. To create a new agent group, click Add.

Configure a prompt action for your agents

A Prompt Action allows the orchestrator agent to perform a reasoning or transformation step using the AI model without calling an external API. Instead, the action relies entirely on prompt instructions to guide how the model should process information, make decisions, or generate intermediate outputs during an agentic RAG workflow.

To configure the a prompt action for the orchestrator agent, expand the Orchestrator section and specify the following:

1. To edit an existing prompt action, click > Edit. To create a new prompt action, click + Prompt Action.

2. In the Name field, specify the name for your prompt action.

3. In the Description field, provide the instructions that generate a response by the AI model. In the Orchestrator agent, this field will determine what type of response will be generated. For example, if it is a generic question or a question that relies on crawled content.

4. In the Response Description field, provide the instructions for the AI model to transform how it will respond to the user's query.

   If you are defining a Response Generator action, you can include parameters in the Description and Response Description fields that were collected from the Query Generator agent action to support your AI model's response. You can denote these parameters by including them inside curly brackets "{}". For example, "Extract and format filter criteria from the user input. Build filters only if they match these available properties with their values: {suggestedFilters}".

5. Click Save.

Configuring an API Action for your agents

An API Action defines how the orchestrator agent can call an external API during agentic RAG execution. This allows the AI model to retrieve live data, trigger workflows, or enrich responses as part of conversational search. Each field you configure teaches the AI what the API does, how to call it, and how to interpret the response.

To configure an API action for the orchestrator agent, expand the Orchestrator section and specify the following:

1. To edit an existing API action, click > Edit. To create a new API action, click + API Action.

2. In the Name field, specify the name for your API action.

3. In the Description field, provide the instructions that generate a response by the AI model, explaining when and why the AI should use this API. For example, let's imagine you are configuring an agent action that creates a ticket in Jira. your description may be "Creates a ticket in Jira".

4. In the Parameters field, click + Add Parameter and define the inputs that the API expects to tell the AI what information to collect.

   1. In the Name field, enter the name of the parameter(s) that you want the AI to collect from the endpoint. For example, if we are configuring an agent that creates a ticket in Jira, the names may include title, description, issueType, etc.

   2. In the Description field, provide a description of the parameter. Using the example, above, for the issueType parameter, enter "The issue type of the ticket. For example, Task, Bug, Story."

5. In the Response Description field, provide a description to explain what the API returns and how the AI should use the data. For example, "Returns the employee’s name, department, and job title.".

   If you are defining a Response Generator action, you can include parameters in the Description and Response Description fields that were collected from the Query Generator agent action to support your AI model's response. You can denote these parameters by including them inside curly brackets "{}". For example, "Extract and format filter criteria from the user input. Build filters only if they match these available properties with their values: {suggestedFilters}".

6. Expand the API Details section and provide the following:

   1. In the API Endpoint field, specify the API URL that SmartHub will call when the agent invokes this action. For example, https://api.company.com/employees.

   2. In the Request Headers field, click + Add Header or + Add Secure Header to add request headers to your API request. For example, Authorization: Bearer {{oauth_token}}, Content-Type: application/json, etc.

   3. In the Request Type field, specify the HTTP method your agent uses. The possible options include:

      • GET

      • POST

      • PUT

      • DELETE

   4. In the Request Body field, enter the JSON payload sent with the request. For example, the Jira ticket creation use case we established previously may look like the following:

      {
          "fields": {
              "description": {
                  "content": [
                      {
                          "content": [
                              {
                                  "text": "{description}",
                                  "type": "text"
                              }
                          ],
                          "type": "paragraph"
                      }
                  ],
                  "type": "doc",
                  "version": 1
              },
              "issuetype": {
                  "name": "{issueType}"
              },
              "project": {
                  "key": "GG"
              },
              "summary": "{title}"
          }
      }

   5. In the Response Key Path field, Specify where the relevant data exists in the API’s JSON response. For example:

      {
        "responseKeyPaths": {
          "title": "fields.summary",
          "issueType": "fields.issuetype.name",
          "description": "fields.description.text"
        }
      }

7. Expand the Authentication Details section and provide the following:

   1. In the Authentication Type field, Specify the authentication you want to use, either None or OAuth.

   2. In the OAuth Configuration field, specify the OAuth provider that you configured on the API Access page.

8. Click Save.

Understanding pre-populated agent actions

By default, The Agents management page comes pre-populated with specific agent actions. The following explains each default action and how you can configure them for your needs.

Orchestrator agent

Retrieve Answer for Standalone Request

This action uses the user's search query to perform a contextual search and retrieve an informed answer. To edit the configuration of this action, do the following:

1. On the Agents Management page, click > Edit.

2. In the Name field, you can alter the name of the action.

3. In the Description field, you can alter the description that you will send to the AI resource to generate a response. By default, this field is set to "Generate a response for the user questions about weather, telling a joke, writing a description, history, calculating sums, and other general questions."

4. In the Response Description field, you can alter the description for who the response is intended for. By default, this field is set to "The user request".

Download Required

This action is used to generate a downloadable document based on the user's conversational search history in .pdf or .docx format. For example, after chatting with the conversational search bot, when the user asks a question, such as "provide me a downloadable document on everything we've discussed so far", the bot will respond with a link to download the generated document.

1. On the Agents Management page, click > Edit.

2. In the Name field, you can alter the name of the action.

3. In the Description field, you can alter the description that you will send to the AI resource to generate a downloadable document. By default, this field is set to "Check if the user is asking for a download based on the chat history."

4. In the Response Description field, you can alter the description for who the response is intended for. By default, this field is set to "User question".

General affairs

This action is used to generate a response when the AI does not need to reference your indexed content.

1. On the Agents Management page, click > Edit.

2. In the Name field, you can alter the name of the action.

3. In the Description field, you can alter the description that you will send to the AI resource to generate a response. By default, this field is set to "Process the user's query or request by performing a contextual search to retrieve ."

4. In the Response Description field, you can alter the description for who the response is intended for. By default, this field is set to "The user request".

Query Processor agent

Determine Sources

This action is used to analyze the user's query and determine which sources are relevant to it. This helps to minimize the searched content, reducing the time it takes to generate a response.

1. On the Agents Management page, click > Edit.

2. In the Name field, you can alter the name of the action.

3. In the Description field, you can alter the description that you will send to the AI resource to determine which sources are relevant. By default, this field includes rules for the AI, a sample JSON for source definition, and query/output formatting.

4. In the Parameter field, you can alter the parameter and description of the inputs that the API uses to collect the source. By default, this is set to Sources: An array of determined sources from the user's input. This parameter can be used later to specify additional query template constraints in your conversational search settings.

Determine Filters

This action is used to extract and filter criteria based on the user's search query and only retrieve a response that is relevant to the filtered query.

1. On the Agents Management page, click > Edit.

2. In the Name field, you can alter the name of the action.

3. In the Description field, you can alter the description that you will send to the AI resource to determine which filters are relevant. By default, this field includes rules for the AI and sample formats for different filters.

4. In the Parameter field, you can alter the parameter and description of the inputs that the API uses to collect the filters. By default, this is set to computedFilter: Represents the filter.

Determine Instructions

This action is used to extract all of the instructions provided in the user input and chat history to determine how the response should be generated.

1. On the Agents Management page, click > Edit.

2. In the Name field, you can alter the name of the action.

3. In the Description field, you can alter the description that you will send to the AI resource to determine the instructions to use. By default, this field is set to "Extracts all instructions provided in the user input and the chat history that defines how the answer should be generated. This function captures guidance on presentation, tone, style, or any other preferences, ensuring the final answer aligns with the user's specified directives".

4. In the Parameter field, you can alter the parameter and description of the inputs that the API uses to collect the instructions. By default, this is set to instructions: The set of instructions.

Determine Search Query

This action is used to clean the user's search input to best determine the search query to run after stripping out any unnecessary terms such as stop words or keywords determined by the previous actions.

1. On the Agents Management page, click > Edit.

2. In the Name field, you can alter the name of the action.

3. In the Description field, you can alter the description that you will send to the AI resource to determine the search query to run. By default, this field is set to "Extracts the important keywords from a user input intended for search by stripping out filters, sources, instructions, and redundant stop words. Returns only the essential terms required to execute a targeted search to get the answer. Only remove keywords that you are positive are not necessary for a search".

4. In the Parameter field, you can alter the parameter and description of the inputs that the API uses to collect the search query. By default, this is set to cleanSearchQuery: The search query to run.

Response Generator agent

Generate Answer

This action provides the instructions to the search bot to generate an answer to the user's query.

1. On the Agents Management page, click > Edit.

2. In the Name field, you can alter the name of the action.

3. In the Description field, you can alter the description that you will send to the AI resource to determine how the answer will be generated. By default, this field is set to "You are a chatbot engaged in a conversation with a human. When presented with a question and a context composed of parts from multiple documents, provide a concise answer solely based on the provided context. If an answer cannot be derived from the context, respond with: 'Sorry, I don't have the information needed to answer your question'. When you can provide an answer, state it concisely at the beginning of your response without any prefix or any other special character and then include the list of the sources. Return all the source values that you found in 'clickUri' property as a list in this format: <SourceDocsURL>URL1;URL2;URL3;etc.</SourceDocsURL>". Additionally, you can include any parameters that you specified in the Query Processor agent. For example, you can add {instructions} to append the instructions information.

Result Insight agent

Suggest Filters

This action allows the AI to suggest filters for the user's query in the response.

1. On the Agents Management page, click > Edit.

2. In the Name field, you can alter the name of the action.

3. In the Description field, you can alter the description that you will send to the AI resource to determine how the filters will be generated. By default, this field is set to "If ChunkData (search results or retrieved context) is present in the chat, suggest filters in the format PropertyName1:'PropertyValue1';(PropertyName2:'PropVal2' OR PropertyName2:'PropVal3'); etc. only from the ChunkData content. Do not suggest filter which are not from the ChunkData content".

4. In the Parameter field, you can alter the parameter and description of the inputs that the API uses to collect the suggested filters. By default, this is set to suggestedFilters: The suggested filters calculated from the response.

Suggest Followup Questions

This action allows the AI to suggest followup questions for the user's query in the response.

1. On the Agents Management page, click > Edit.

2. In the Name field, you can alter the name of the action.

3. In the Description field, you can alter the description that you will send to the AI resource to determine how the followup questions will be generated. By default, this field is set to "If ChunkData (search results or retrieved context) is present, analyze its content carefully. Based on the key topics, insights, or gaps in the retrieved information, generate three relevant followup questions that a curious or goal-oriented user might naturally ask next".

4. In the Response Description field, you can alter the how the followup questions are formatted and displayed. By default, this field is set to "A list of three well-crafted, contextually relevant followup questions derived from the information in ChunkData. Each question separated by a semi-colon".