openai/openai-realtime-agents

Features

• Realtime AI Session Managers - Builds conversational agents using the OpenAI Realtime API with support for switching between models and roles.
• Multi-Agent Routing Systems - Directs users between different specialized AI agents based on their intent to handle complex task requirements.
• Agent Routing Frameworks - Provides a framework for directing user requests to specialized agents using a defined intent graph.
• Tool Call Execution Loops - Processes model-generated function requests by executing local code and feeding results back into the context.
• Hybrid LLM Task Distributors - Implements a pattern for splitting workloads between fast conversational models and high-intelligence models.
• Hybrid Model Task Distribution - Implements a hybrid task distributor to split workloads between fast conversational models and high-intelligence models.
• Multi-Agent Orchestration Systems - Routes user requests between specialized agents using a graph to manage complex task requirements.
• Full-Duplex Multimodal Interaction - Maintains a persistent open connection for simultaneous audio and text streaming between client and server.
• Realtime API Agent Frameworks - Provides a framework for building voice and text agents using the OpenAI Realtime API with agentic patterns.
• Tiered Model Workload Splitting - Distributes workload between a low-latency model for interaction and a high-reasoning model for complex tool execution.
• Intent-Based Routing - Directs user requests between specialized model instances based on a predefined map of intents.
• Conversational State Managers - Guides a voice AI through a strict sequence of steps to collect and verify user information.
• Finite State Machine Managers - Enforces strict sequences of steps to collect and verify data points during model conversations using a finite state machine.
• Sequential State Enforcement - Guides models through a strict sequence of steps to collect and verify specific data points sequentially.
• Streaming Output Modifiers - Intercepts model responses in the stream to validate content against safety and compliance rules.
• AI Output Safety Filters - Scans generated voice and text responses against safety and compliance rules before they reach the end user.
• Model Safety Filters - Scans generated model responses against safety and compliance rules in real-time to validate content.
• Conversational State Machines - Guides model behavior through a defined sequence of states to ensure structured data collection.

Star history