Learning: OpenAI Agent SDK - Conditional tool enabling patterns & Run context

on
https://www.pexels.com/photo/fit-runner-standing-on-racetrack-in-athletics-arena-3756165/
https://www.pexels.com/photo/fit-runner-standing-on-racetrack-in-athletics-arena-3756165/


This is the second blob (following the first one) on OpenAI SDK, and it is about Conditional tool enabling patterns & Run Context. We continue to use the healthcare domain to illustrate how these work together.

This blob outlines the implementation of conditional tool enabling patterns in agentic systems, demonstrating how context-driven tool availability can create dynamic, adaptable workflows using the OpenAI Agent SDK with Azure AI Foundry (using Azure OpenAI model) integration.

view documentation | source code

Overview

Conditional tool enabling patterns allow agents to dynamically control which tools are available based on runtime context, user preferences, or system state. This approach is crucial for building adaptive systems that can modify their behavior and capabilities based on changing requirements or configurations.

In user preference systems, conditional enabling allows customization of available features based on user settings or subscription levels. For resource management, this pattern enables efficient allocation by activating only necessary tools based on current system load or capacity. Compliance scenarios benefit from conditional enabling by restricting tool availability based on regulatory requirements or security clearances. Additionally, conditional patterns facilitate progressive disclosure by gradually exposing more advanced features as users demonstrate proficiency or meet specific criteria.

Key Concepts

Context-Driven Enabling

The mechanism by which tool availability is determined based on runtime context, user preferences, or system state rather than static configuration.

Conditional Functions

Predicate functions that evaluate context and return boolean values to determine whether specific tools should be enabled for a given execution.

Run Context Wrapper

A container that carries contextual information throughout the agent execution lifecycle, enabling consistent context-aware decisions.

Dynamic Tool Registry

The system that manages which tools are available to agents based on conditional evaluation at runtime.

Architecture

The following diagram illustrates the conditional tool enabling architecture:

The user can choose the languages that the user wishes to subscribe to. And then the necessary language tools will be enabled for LLM consumption.

Flow Description

  1. User Input: Medical message with language preference selection
  2. Context Selection: User chooses from predefined language combinations
  3. Context Evaluation: Conditional functions determine which tools to enable
  4. Dynamic Execution: Only enabled tools are available to the orchestrator agent
  5. Multi-Language Output: All enabled translation agents execute in parallel

Code Structure

The conditional tool enabling implementation consists of:

  • Context Model: Pydantic model defining the runtime context structure
  • Conditional Functions: Predicate functions that evaluate context for tool enabling
  • Dynamic Tool Registry: Tools with conditional enabling based on runtime context
  • Context Selection: User interface for choosing execution context
  • Structured Output: Consistent output format across all enabled tools

Usage

Running the Example

```bash
# Activate virtual environment
source .venv/bin/activate

# Run the conditional tool enabling example
python -m openai_agent.agentic_patterns.tool_conditional
```

Demo (please view the video in full screen and it does not have audio)



 


Comments

Post a Comment