Azure/azure-sdk-for-java com.azure+azure-ai-agents_2.6.0

on GitHub

2.6.0 (2026-09-22)

Features Added

• Added preview voice-agent lifecycle support through VoiceAgentDefinition, AgentKind.VOICE, and
  AgentEndpointProtocol.VOICE. With AgentsClientBuilder.allowPreview(true), voice agents can be created and versioned
  with AgentsClient and AgentsAsyncClient, with configuration for speech models, instructions, structured inputs,
  audio and transcription, turn detection, greetings and interim responses, function/MCP/toolbox/system tools,
  subagents, persistence, avatars, and animation.
• Added createAgentFromPrompt and createAgentFromPromptWithResponse to BetaAgentsClient and
  BetaAgentsAsyncClient for generating and creating agents from kind-specific high-level inputs.
• Added preview BetaVoiceAgentWebSocketClient, BetaVoiceAgentWebSocketAsyncClient,
  BetaVoiceAgentWebSocketSessionClient, and BetaVoiceAgentWebSocketSessionAsyncClient for authenticated realtime
  wss:// sessions. The clients support typed and raw events, text and audio input, response cancellation,
  client-executed function results, persisted conversations, agent-version overrides, and the related realtime session,
  conversation-item, response, transcription, MCP, audio, avatar, and RTC event models.
• Added VoiceAgentWebSocketConnectionOptions for session IDs, structured inputs, persistence, version selection,
  handshake and close timeouts, receive-buffer and message-size limits (32 MiB by default), overflow policies, and
  opt-in recovery from malformed events. Sessions also support raw JSON sends, complete unknown-event payloads,
  UTF-8 binary JSON, per-event synchronous receive timeouts, and custom close codes and reasons.
• Added type-safe constructors and accessors for union-valued voice and realtime properties, including session
  configurations, tool choices, token limits, tracing, truncation, voices, MCP options, and transcription usage.
• Added preview BetaVoiceAgentsConversationsClient and BetaVoiceAgentsConversationsAsyncClient for listing,
  retrieving, and deleting persisted voice-agent conversations; reading responses and conversation items; and retrieving
  or downloading input, generated, and full-conversation audio.
• Added preview BetaVoiceAgentsTelephonyClient and BetaVoiceAgentsTelephonyAsyncClient for Teams Phone Extension and
  Twilio bindings, call listing and inspection, call transfer and termination, transfer-target management, and scheduled
  outbound call jobs with retry and cancellation support. All voice beta clients are built through
  AgentsClientBuilder.beta() and automatically add the VoiceAgents=V1Preview feature opt-in.
• Added preview GitHub Copilot harness and built-in toolset support for prompt agents through
  PromptAgentDefinition.setHarness(...), GitHubCopilotHarness, GitHubCopilotToolsetPreview, and related types.
  Prompt agents can also reference versioned Foundry skills through PromptAgentDefinition.setSkills(...) and
  SkillReference.
• Added preview Model Router session affinity. AzureCreateResponseOptions.setRoutingConfig(...) configures affinity
  mode and session ID, while ModelRouterDetails.getSessionAffinity() exposes the effective mode, source, and routing
  decision.
• Added invocation-protocol content moderation through RaiConfig.setInvocationsModeration(...), with JSON/text input
  and output selectors plus SSE event selectors for streaming responses.
• Added invokeLatestToolboxMcp and invokeLatestToolboxMcpWithResponse to ToolboxesClient and
  ToolboxesAsyncClient. ToolboxDetails now exposes its last-updated time and latest-version details through
  getUpdatedAt() and getVersions().
• Added AgentDetails.getConfigurationState() to expose an agent's administrative enablement state and
  AgentSessionResource.getStoppedAt() to expose when a hosted-agent session last stopped or became idle.
• Added ToolType.GITHUB_COPILOT_TOOLSET_PREVIEW and ToolType.BROWSER_AUTOMATION discriminator values, along with
  typed MCP list-tools models for voice-agent realtime sessions.
• Added VersionSelector.setVersionSelectionRule(...) as a convenience for configuring one version-selection rule.
• Added public StreamingResponseUtils in com.azure.ai.agents.util for converting OpenAI streaming responses to Azure
  SDK IterableStream and Reactor Flux types.
• Added voice-agent samples covering lifecycle and version management, guided generation, audio and tool configuration,
  persisted conversations and audio, synchronous and asynchronous live text, asynchronous Java Sound audio with
  barge-in, and client-executed function tools.

Bugs Fixed

• Fixed serialization of openai-java Reasoning values in PromptAgentDefinition so retrieved definitions can be
  reused to create new agent versions without emitting unsupported request properties.
• Fixed polling for agent-optimization jobs that return the cancelled status spelling.

Other Changes

• Updated existing response, streaming, and tool samples to configure agent endpoints and route requests through
  buildAgentScopedOpenAIClient(...) or buildAgentScopedOpenAIAsyncClient(...).
• Added Reactor Netty, Netty, OkHttp, and Okio runtime dependencies for voice-agent WebSocket transport support.
• Regenerated the client from the updated TypeSpec specification.