v2.0.0

Release Notes - v2.0.0

🎯 Major Features

Scenario Builder Enhancements

• Visual graph-based Scenario Builder with drag-and-drop functionality and panning controls
• Pre-built handoff condition patterns (Authentication, Fraud/Security, Escalation, Technical Support)
• Unified HandoffService consolidating handoff logic across all orchestrators
• Support for orchestration.yaml naming convention for scenario definitions

Voice Handler & Architecture Refactoring

• Complete VoiceHandler refactoring with unified lifecycle management
• Migrated MediaHandler to VoiceHandler with proper lifecycle controls
• Cleaned up lifecycle management logic into dedicated structure
• Voice Live readiness status now uses event envelope format

Evaluation & Testing Framework

• New model evaluation playground with A/B testing capabilities
• Comprehensive test suite for VoiceLive handler, cascade orchestrator, and DTMF processor
• Scenario orchestration contract tests to preserve functional contracts during refactoring
• Enhanced load testing with Locust scripts and mkdocs-mermaid-zoom support

Agent Builder & Multi-Model Support

• Enhanced Agent Builder with multi-model configuration and voice improvements
• Dual model configuration support (GPT-4o and GPT-4.1)
• Responses API infrastructure for advanced model routing
• Session performance monitoring dashboard

🔧 Enhancements

Telemetry & Observability

• OpenTelemetry consolidation with proper span hierarchy
• Shared metrics factory for lazy initialization
• Enhanced tracing for LLM, Speech, and ACS calls
• Comprehensive logging improvements across connection manager, warmable pool, Redis, and speech modules

TTS & Speech Processing

• Text sanitization and sentence boundary detection for improved audio quality
• Enhanced audio transcription settings with configurable parameters per agent
• Improved TTS playback and voice handling
• Agent voice prioritization over defaults for initial/return greetings

Deployment & Infrastructure

• Enhanced deployment scripts with pre/post-provisioning hooks
• Azure CLI extension checks and EasyAuth configuration automation
• Dev Container testing for AZD hooks with environment validation
• Reusable CI/CD workflow templates with parallel testing across Linux/macOS/Windows
• GitHub PAT support with enhanced environment variable handling
• Troubleshooting guidance for Terraform state lock errors

Frontend UI

• Disabled view toggle buttons for chat/graph/timeline in ConversationControls
• Enhanced ProfileDetailsPanel with improved styling
• Session management improvements with better context handling
• Backend health monitoring and status indicators

Configuration & State Management

• .env.local support for configuration overrides
• App configuration bootstrap for environment variable initialization
• Improved configuration reliability and dynamic loading
• Azure App Configuration module with RBAC and Key Vault integration

🐛 Bug Fixes

• Fixed Redis connection handling with proper error recovery
• Eliminated duplicate UI updates in LiveOrchestrator during context-only session updates
• Corrected pull_request event handling in Azure deployment workflow
• Fixed foundry account and project naming conventions for consistency
• Updated naming from rtaudioagent to artaudioagent in environment files
• Fixed health check endpoint in postprovision script to use correct API path
• Resolved duplicate handoff calls in fraud detection template
• Fixed image paths in quickstart guide for accurate rendering

🗑️ Removals

• Removed legacy Redis management classes and related files
• Removed deprecated latency tools (replaced by OpenTelemetry metrics)
• Removed backend IP restrictions configuration
• Removed unused backend API public URL variables
• Cleaned up outdated troubleshooting documentation

📚 Documentation

• Updated quickstart guide with demo profile creation and screenshots
• Added agent builder template selection screenshots
• Enhanced scenario testing instructions for clarity
• Improved README layout and navigation links
• Added deployment badges and workflow documentation
• Added troubleshooting steps for common deployment issues

🔐 Security

• Dependency updates: vite (6.3.5 → 6.4.1), js-yaml (4.1.0 → 4.1.1)
• Updated on-headers and serve packages
• Enhanced EasyAuth configuration and integration

---

This release represents significant improvements to the Scenario Builder, voice handling architecture, and deployment infrastructure, with a strong focus on testing, observability, and developer experience.

V2.0.0.beta

Beta release featuring the YAML-driven agent system, multi-scenario orchestration, and Azure VoiceLive SDK integration. This release represents a complete architectural evolution from v1.x.

Added

• Unified Agent Framework — YAML-driven agent definitions (agent.yaml) with Jinja2 prompt templating and hot-reload
• Scenario Orchestration — Multi-agent scenarios with orchestration.yaml defining agent graphs, handoffs, and routing
• Azure VoiceLive SDK — Native integration with gpt-4o-realtime for ~200ms voice-to-voice latency
• Industry Scenarios — Banking (concierge, fraud, investment) and Insurance (FNOL, policy advisor, auth) ready-to-use
• 15+ Business Tools — Authentication, fraud detection, knowledge search, account lookup, card recommendations
• Streaming Mode Selector — Frontend toggle between SpeechCascade and VoiceLive orchestrators
• Profile Details Panel — Real-time caller context display with tool execution visualization
• Demo Scenarios Widget — One-click scenario switching for demos and testing

Enhanced

• Package Management — Migrated to uv for 10x faster installs with reproducible uv.lock
• OpenTelemetry — Full distributed tracing across LLM, Speech, and ACS with latency metrics
• Phrase Biasing — Dynamic per-agent phrase lists for improved domain-specific recognition
• Agent Handoffs — Seamless context preservation during multi-agent transfers
• Devcontainer — ARM64/x86 multi-arch support with optimized startup

Fixed

• VoiceLive "already has active response" conflicts during rapid handoffs
• LLM streaming timeouts (now 90s overall, 5s per-chunk with graceful cancellation)
• Tool call index validation filtering malformed responses
• Docker build optimization removing unnecessary apt upgrades

v1.0.0

v1.0.0 Release of the Azure Real Time Agent Accelerator

Custom Speech Cascade multi-agent orchestration + Azure Communication Services
VoiceLive API implementation, single agent
Sample scenarios modeled around insurance, with composable tools for customizations for other industries
End-to-End deployment via azd up