Azure/azure-sdk-for-java com.azure+azure-ai-agents_2.9.0

on GitHub

2.9.0 (2026-10-30)

Features Added

• Added hosted-agent memory snapshot opt-in through HostedAgentDefinition.setMemorySnapshotEnabled(...) and
  isMemorySnapshotEnabled(). Sessions for a snapshot-enabled agent version start from a snapshot of its initialized
  process instead of starting the process from its image.
• Added preview voice-agent soundscapes through VoiceAgentDefinition.setSoundscape(...),
  VoiceAgentSoundscapeConfiguration, and related track and built-in sound models. Soundscapes support background
  ambience, delayed working audio while tools are pending, and opening, transfer, and closing cues. Rendering is
  currently limited to telephony calls; persisted conversation recordings do not include the soundscape mix.
• Added preview input-transcript profanity filtering through VoiceAgentInputTranscription.setProfanityFilter(...)
  and VoiceAgentProfanityFilterMode, with MASKED, REMOVED, and UNFILTERED modes. Filtering is supported with
  managed Azure Speech, including Custom Speech, and must be configured before the session starts.
• Added VoiceAgentInputTranscriptionModel.AZURE_SPEECH_STREAMING for preview streaming Azure Speech transcription
  in cascaded text-model conversations. It requires a nonempty language setting and supports Custom Speech,
  phrase hints, and profanity filtering; it is not supported with realtime speech-to-speech models.
• Added preview VoiceAgentNoiseReductionType.AZURE_DEEP_NOISE_SUPPRESSION_HEADSET and
  AZURE_PERSONAL_NOISE_SUPPRESSION modes for local voice activity detection. These modes enhance detection audio
  without changing the audio used for recognition.
• Added preview AgentEndpointProtocol.VOICE_CONVERSATION_ENGINE for the hosted-agent text conversation protocol
  used by Voice Live Bridge. Audio transport remains outside this protocol.

Breaking Changes

• Updated the preview hosted voice conversation-engine contract: hosted agents referenced by
  VoiceHostedAgentConversationEngine must declare voice_conversation_engine version 1.0.0.
  Declaring invocations_ws is optional and no longer substitutes for the dedicated voice conversation-engine protocol.

Bugs Fixed

• Fixed cursor-based list operations that stopped after the first page, including memory-store, memory-item,
  toolbox, and voice-agent listings. Paging now follows the service's last_id cursor and supports resuming from
  a continuation token.
• Fixed cursor-based paging to stop when the service returns has_more: false, even if the response still includes
  a last_id, avoiding unnecessary requests after the final page.

Other Changes

• Deprecated BrowserAutomationPreviewTool, BrowserAutomationPreviewToolboxTool, and the corresponding
  BROWSER_AUTOMATION_PREVIEW tool-type values. Use BrowserAutomationTool, BrowserAutomationToolboxTool,
  and the BROWSER_AUTOMATION values instead. See Browser Automation documentation.
• Regenerated the client from the updated Foundry TypeSpec specification.
• Updated azure-core from 1.60.0 to 1.61.0 and azure-core-http-netty from 1.16.8 to 1.17.0.
• Updated Reactor Netty from 1.2.18 to 1.3.7 and Netty from 4.1.138.Final to 4.2.19.Final,
  including the Netty codec module changes required by the newer runtime.