SearchAI Agents

Overview

Agents is a SearchAI feature that enables intelligent query execution using AI language models over your indexed search collections. Primary applications include automated question answering, complex multi-step reasoning, scheduled content analysis, and multi-agent orchestration.

Key Features

Create and manage LLM and Workflow agents
Bind agents to one or more search collections.
Invoke agents interactively through a chat interface
Orchestrate multiple agents simultaneously with aggregated results
Schedule automated agent runs using cron-based expressions
Configure webhook and email notifications for scheduled executions

Types of Agents

LLM: Answers queries by retrieving relevant content from search collections and generating a response using the configured language model.
Workflow: Handles complex queries by reasoning step by step and calling tools as needed to arrive at a final answer.

Agents List

Agents Table: Displays all agents with columns: No., Type, Name, Mode, Origin, Status, Actions.
Search Bar: Filter agents by name in real time
Create Button: Navigate to the Create Agent form
Refresh Button: Reload the agents list
Orchestrate Button: Open the multi-agent orchestration panel
Invoke (icon): Open the single-agent chat modal
Clone (icon): Duplicate an agent with a new name
Delete (icon): Remove a custom agent (base agents are protected)

Note: There are a total 43 Base Agents available.

Dashboard: In agents dashboard, we can see all the default agents .

Creating SearchAI Agents :

Log in to the Admin Console:

Start by signing in to your SearchBlox Admin Console using your administrator credentials.

Go to the SearchAI tab:

In the dashboard, open the SearchAI section.
Press the “+” button to add a new agent.

Configure the agent details

Name the agent: Provide a meaningful and unique name, such as “Intranet agent”, so it’s easy to identify later.
Description of the agent: Provide a clear Description to explain the purpose and functionality of the agent.
Type: Select the Type of agent based on your use case:
1. LLM: Answers queries by retrieving relevant content from search collections and generating a response using the configured language model.
2. Workflow: Handles complex queries by reasoning step by step and calling tools as needed to arrive at a final answer.
Base agent: select a Base Agent to use an existing configuration as a template.
System Instruction: Define the System Instruction to control how the agent behaves, responds, and interacts with user queries. This instruction acts as the guiding logic for the agent’s responses.
Select Collections: In Select Collections, choose the collections the agent will use to execute queries or tasks. You can select specific collections or use Select All to include all available collections.

Model Settings: Under Model Settings, configure how the agent processes and generates responses:

Provider: Select the LLM provider to power the agent. The Provider to use: Ollama, OpenAI, Google, Azure, Anthropic, or Cohere
Model: Choose the specific model offered by the selected provider.
Temperature: Controls response creativity, lower values produce consistent answers, higher values produce varied responses
RAG Result Limit: Define the number of relevant results to retrieve and use for response generation.

Save the agent:

After confirming all settings, click Save to create your new agent and proceed with further configuration to use it.
Note: Base agents cannot be deleted. Use the Clone action to create a custom copy.

Agent Settings

System Instruction - Defines the agent's behavior and response style
Provider - The Provider to use: Ollama, OpenAI, Google, Azure, Anthropic, or Cohere
Model - The specific model from the selected provider
Temperature: Controls how creative the responses are. Lower values give more consistent answers, higher values give more varied responses.
Execution Mode - Auto lets the agent decide when to use tools. Tools mode forces a fixed set of tools to be used every time.
RAG Result Limit - Maximum number of search results passed to the model as context (0–100)
Include KG Facts - Includes Knowledge Graph facts in the agent context
Max Iterations - (Workflow only) Maximum reasoning steps before the agent returns a response.

Agent Invoke

Click the Invoke icon next to the agent in the list
Select the Collections to query
Enter your query and press Send

Multi-Agent Orchestration:

Log in to your SearchBlox administration panel.
Navigate to the Agents section under SearchAI in the left sidebar.
Select the Multi-Agent Orchestration to combine two or more agents to provide the combined results.

Using Multi-Agent Orchestration:

In Agents, select the agents you want to include in the orchestration.
Mode: Defines how multiple agents are executed during orchestration
- Parallel: Runs all agents simultaneously and aggregates results
- Pipeline: Executes agents sequentially, passing output from one to the next
- Voting: Selects the final response based on majority agreement among agents
The Output setting defines how responses from multiple agents are combined into a single result.
- Concatenate: Combines responses from all agents sequentially into a single output.
- First Success: Returns the first successful response generated by any agent.
- Best: Selects the most relevant or highest-quality response among all agent outputs.
- Merge Unique: Merges responses while removing duplicate or overlapping information.
- Vote Majority: Chooses the response agreed upon by the majority of agents.
- Summarize: Generates a concise summary by combining key points from all agent responses.
Use Collections to specify the data sources that the orchestrated agents can access while processing the query.
The Orchestrator panel allows you to enter a query that will be processed by all selected agents. The system coordinates their responses based on the chosen mode and aggregation settings to generate a unified output.

Schedules

The Schedules tab within an agent allows you to automate agent execution on a defined schedule.

Creating a Schedule

Follow the steps below to set up an automated schedule for your agent to run queries at defined intervals:

Click the Create Schedule button
Enter the schedule Name, Description, and Query
Select the Collections to be queried (visible for base agents)
Choose a Frequency template and set the Start Date and Time
Configure Notifications if required
Set Max Runs and Expiration Date (optional)
Click Save

Notification Settings

Notifications can be sent after each scheduled run. Select a notification type:

None - No notifications sent
Webhook - Sends a POST request to the configured URL after execution
Email - Sends a report to the specified recipients
Both - Sends both webhook and email notifications

Webhook Configuration:

URL - The endpoint to receive the notification payload
Authorization Header - Optional authentication header for the webhook

Email Configuration:

Recipients - Comma-separated list of email addresses
Report Frequency - Daily, Weekly, or Monthly

Schedule List

All schedules associated with the agent are displayed and grouped by status: Active, Paused, Completed, and Failed.

Schedule Actions

Use the action icons on each schedule card to manage it:

View - Opens a read-only summary of the schedule including execution statistics
Edit - Modify the schedule's configuration
Pause / Resume - Temporarily stop or re-enable a schedule
Run Now - Trigger an immediate execution outside the defined schedule
Delete - Permanently remove the schedule

Schedule Info

Click View on any schedule to open its detail page. This displays:

Execution statistics: Total Runs, Successes, and Failures
Schedule details: Name, Agent, Query, Frequency, and Next Run time
Notification configuration summary
Collections bound to the schedule