Skip to main content

Assistant Agent

Overview​

The Assistant Agent is an LLM-powered agent that understands user input, reasons over conversation context, and generates intelligent responses.

It supports no-code configuration, allowing you to attach model clients, guardrails, tools, workbenches, and memory to control behavior.

note
  • Tools are executed via workbenches (Static or MCP-enabled)
  • Custom tools can be written in Python directly from the configuration UI

Adding the Assistant Agent​

Actions​

  • Open Team Builder
  • Ensure the team type is set to Agent
  • Drag and drop Assistant Agent into the canvas

Attaching a Model Client​

Actions​

  • Drag a Model Client (e.g., Azure OpenAI)
  • Configure credentials and deployment
  • Connect it to the Assistant Agent

Notes​

  • A Model Client is mandatory
  • Only one model client can be attached

Enabling Guardrails (Optional)​

Actions​

  • Drag a Guardrail Component into the agent
  • Configure entities to mask (optional)
  • Enable Unmask LLM Output if required

Behavior​

  • User input is always masked before LLM execution
  • Output masking is optional
  • Original values are restored based on configuration

Configuring Tools & Workbenches​

Tools are executed through workbenches, which act as the agent’s runtime environment for tools.

Supported workbench types:

  • Static Workbench – Local Python tools
  • MCP-enabled Workbench – Remote / managed execution

Adding Custom Tools to an Agent​

Step 1: Add a Static Workbench​

  • Open Assistant Agent → Configuration
  • Click Add Workbench
  • Choose Static Workbench

Step 2: Open the Workbench Editor​

  • Open the created Static Workbench
  • Click Edit

Step 3: Add a Tool​

  • Click Add Tool
  • Provide the Python code and required configuration

Each tool typically defines:

  • Tool name
  • Input arguments schema
  • Executable Python logic

Once saved, the tool becomes immediately available to the agent.

note
  • Multiple tools can exist in a single workbench
  • Multiple workbenches can be attached to one agent
  • Supports iterative tool calls, streaming output, and error handling
  • MCP workbenches enable remote and managed execution

Configuring Agent Behavior​

Settings​

  • Agent name and description
  • System prompt (LLM instructions)
  • Streaming response mode

Capabilities and Use Cases​

Key Capabilities​

  • LLM-based reasoning and responses
  • Context-aware conversations
  • Secure PII handling via guardrails
  • Static and MCP workbench integration
  • Custom Python tool execution
  • Streaming and non-streaming output
  • No-code configuration

Common Use Cases​

  • Conversational assistants
  • Enterprise Q&A bots
  • Secure document and invoice analysis
  • Customer support agents
  • Internal knowledge assistants

Summary​

The Assistant Agent enables powerful LLM-driven interactions with optional guardrails, workbench-based tools, and custom Python integrations, making it suitable for both simple chatbots and complex, tool-driven workflows.