Tool-using agents extend beyond conversation and knowledge retrieval by connecting AI capabilities to external systems, databases, and services. These agents can perform actions in the real world, access live data, and execute complex workflows, turning AI from a passive information source into an active participant in your business processes.

​
What Are Tool-Using Agents?

Tool-using agents combine the language understanding and generation capabilities of AI models with the ability to:
  1. Determine when to use tools based on user requests and context
  2. Select appropriate tools from a set of available capabilities
  3. Format tool parameters correctly for successful execution
  4. Interpret tool results and incorporate them into responses
  5. Manage sequences of tool use for multi-step workflows
Tools represent the bridge between AI language capabilities and functional business systems. They transform agents from conversational interfaces into operational participants in your workflows.

​
Architecture of Tool-Using Agents

The architecture of a tool-using agent involves several key components:

​
Key Components

Language Model

The core reasoning engine that understands requests, determines when tools are needed, and formulates responses

Tool Registry

The collection of available tools with their descriptions, parameters, and capabilities

Tool Selection Logic

Rules and algorithms that determine which tool to use in which situation

Parameter Formation

Process of translating user inputs into structured parameters for tool execution

Execution Engine

System that runs the selected tools with provided parameters and captures results

Result Interpretation

Logic for processing tool outputs and incorporating them into agent responses

​
How Tool-Using Agents Work

1

Request Understanding

The agent analyzes the user request to determine intent and required actions.During this phase, the agent:
  • Identifies the user’s core needs and intent
  • Determines whether tools are required
  • Extracts relevant information for potential tool use
  • Plans a response strategy
2

Tool Selection

If tools are needed, the agent selects the most appropriate tool(s) for the task.Selection criteria include:
  • Tool capabilities compared to task requirements
  • Parameter availability from user input
  • Tool reliability and performance characteristics
  • Authorization and access rights
3

Parameter Preparation

The agent formats the necessary parameters for tool execution.This process involves:
  • Extracting explicit parameters from user input
  • Inferring implicit parameters based on context
  • Formatting parameters according to tool requirements
  • Validating parameters before submission
4

Tool Execution

The selected tool runs with the prepared parameters.During execution:
  • The system handles authentication and authorization
  • The request is sent to the appropriate endpoint
  • Execution is monitored for timeouts or errors
  • Results or errors are captured for further processing
5

Result Interpretation

The agent processes the tool outputs into a usable form.This includes:
  • Parsing structured data from tool responses
  • Extracting relevant information for the user’s request
  • Handling error cases appropriately
  • Determining if additional tool calls are needed
6

Response Generation

Finally, the agent creates a response incorporating tool results.The response typically:
  • Presents tool outputs in a user-friendly format
  • Provides context and interpretation of results
  • Maintains conversation continuity
  • Suggests next steps or additional actions

​
Implementing Tool-Using Agents in Prisme.ai

Prisme.ai provides two complementary approaches to implementing tool-using agents:
Configure tools through the graphical interface in AI Knowledge.Key Features:
  • Graphical tool selection and configuration
  • Pre-built integrations for common services
  • Custom tool definition with visual editors
  • Integration with AI Knowledge RAG capabilities
  • Simplified deployment and management
This approach is ideal for business teams and subject matter experts who need to add tool capabilities to their agents without coding.

​
AI Builder as the Integration Foundation

AI Builder represents the core of Prisme.ai’s tool capabilities, providing:

Event-Driven Architecture

Communication between components via a robust event system

Automations

Server-side workflows that can be triggered by events, APIs, or schedules

API Integration

Built-in capabilities for connecting to external services and systems

Custom Logic

Support for complex conditional processing and business rules

Activity Monitoring

Comprehensive event logging and debugging capabilities

Marketplace Apps

Pre-built integrations that can be installed and configured

​
Available Tool Types

Prisme.ai supports several categories of tools that can be integrated with your agents:

​
Example: Web Browsing Tool

Here’s an example of the Web Browsing tool integration in Prisme.ai: This tool allows agents to:
  • Search the web for current information
  • Access specific websites for reference
  • Verify facts and claims
  • Gather data from public sources
The implementation includes:
  • A clear description for the AI to understand the tool’s purpose
  • Structured parameter definitions for proper function calling
  • Authentication and access control
  • Error handling and result formatting
Tools like Web Browsing connect through AI Builder automations, which manage execution, error handling, and result processing. The event-driven architecture allows seamless communication between the agent and the tool.

​
Tool-Using Agent Capabilities

Data Access

Retrieve and analyze data from diverse sources
  • Query databases and data warehouses
  • Access real-time system metrics
  • Retrieve customer records
  • Generate reports and analytics

System Integration

Interact with enterprise systems and applications
  • Create tickets in ITSM systems
  • Update CRM records
  • Submit expense reports
  • Schedule resources and meetings

Process Automation

Execute multi-step business processes
  • Onboarding workflows
  • Approval processes
  • Document management
  • Multi-system orchestration

Information Services

Connect to external information sources
  • Weather and location services
  • Financial market data
  • Industry news and updates
  • Public data sources

​
Example Use Cases

Purpose: Automate IT service desk functions through system integrationTool Integrations:
  • ITSM system connector for ticket creation and status updates
  • Knowledge base search tool for documentation
  • System diagnostics tools for troubleshooting
  • User directory access for account information
Example Workflow:
  1. User reports an issue with email access
  2. Agent uses directory tool to verify account status
  3. Agent runs diagnostic tools to check service availability
  4. Agent creates a ticket in the ITSM system if needed
  5. Agent provides status and next steps to the user

​
Getting Started with Tool-Using Agents

1

Define Your Requirements

Determine what systems and data your agent needs to access.Key questions to consider:
  • What actions should the agent be able to perform?
  • What data sources are required?
  • What systems need to be integrated?
  • What security and compliance requirements apply?
  • What user workflows will the agent support?
2

Explore Available Tools

Identify which tools can fulfill your requirements.Options include:
  • Built-in tools for common capabilities
  • Marketplace apps for pre-built integrations
  • Custom tool development for specialized needs
Prioritize reuse of existing integrations when possible.
3

Plan Tool Integration

Design how your agent will work with selected tools.Consider:
  • Tool selection criteria and logic
  • Parameter extraction from user inputs
  • Error handling and fallback strategies
  • Tool result presentation and formatting
  • Authentication and authorization requirements
4

Implement and Configure

Set up your tools using AI Knowledge or AI Builder.For simpler needs:
  • Use AI Knowledge’s graphical interface
  • Configure pre-built tools
  • Connect to existing automations
For advanced requirements:
  • Create custom automations in AI Builder
  • Develop specialized integration logic
  • Implement complex workflows and error handling
5

Test and Refine

Validate your agent’s tool usage capabilities.Testing approaches:
  • Function-level testing of individual tools
  • Scenario-based testing of complete workflows
  • Edge case testing for error conditions
  • Performance testing for response times
  • Security testing for access controls
6

Deploy and Monitor

Make your agent available to users and track its performance.Key activities:
  • Define access controls and permissions
  • Set up monitoring and alerting
  • Track usage patterns and success rates
  • Collect user feedback for improvements
  • Monitor for security and compliance

​
Next Steps

Explore these guides to learn more about implementing tool-using agents:

Tool Integration

Learn how to connect agents to external systems and APIs

Tool Selection

Discover strategies for effective tool selection and routing

Memory Management

Understand how to maintain context across tool interactions

Execution & Activity

Monitor and debug tool execution with Activity tracking

Error Handling

Implement robust error management for tool-using agents