js-genai

Environment network configuration now supports "allowlist": "disabled". Also updates the discovery document.

Added lyria-3.5, gemini-omni-1.1-flash, and gemini-omni-flash-preview to the Interactions Model enum, and Interactions now supports continuation_token.

GenerateContent now supports a continuation_token parameter. No other changes are noted in this release.

Gemini 3.8 Flash TTS and Gemini 3.8 Flash Lite TTS models are now available in the SDKs, alongside gcs_metrics_uri support for exporting tuning job metrics and label support in LiveClientSetup and Gemini API GenerateContent. Fixed Trigger.interaction and TriggerCreateParams.interaction to use CreateAgentInteraction.

The Gemini JavaScript SDK adds a Voices API resource and sample_audio to the GAOS SDK, and exposes SpeechMetadata, VoiceConfig.voice, and SpeechAnnotation in public SDKs. Also supports field.value_field syntax for map add_union_value annotations.

Environments files can now be uploaded and downloaded across the Python and TypeScript SDKs, and HttpOptions supports a custom fetch implementation. Also adds a credentials API resource and RetrievalCallStep and RetrievalResultStep to the interactions schema.

Added from_environment support for environment copying. Fixed declaration files by adding @ts-ignore before optional MCP imports, removed an unused field from FunctionResultDelta, and repaired the node fallback manifest and web entries.

The Google Gemini SDK adds support for Gemini 3.8 Flash and updates Flash model descriptions.

Video Understanding support lands in the Interactions API, alongside audio/webm MIME type support and translation_config for GEAP. Also fixed redundant Content-Length headers on file uploads and exposed ProcessingCallStep and ProcessingResultStep in the Interactions SDK.

Model parsing is roughly 2x faster by keeping the Buffer in the Node tokenizer platform. Added AudioTranscriptionConfigMode, environment files support with a scotty file download helper, and TYPE_JPEG2000 to the VideoContent MimeType enum.

The Gemini SDK now supports VERBATIM and SMART modes for audio transcription configs, plus an IDLE state on live connections with REQUIRES_ACTION deprecated. Also adds video resolution/extension task parameters, tool data retention, and a step_count for reinforcement tuning hyperparameters.

Adds support for the gemini-3.7-flash model.

Deferred service tier is now publicly available on Vertex, and LiveServerContent gains an interaction_status field. speech_config is now a structured object.

Adds the Gemini Robotics ER 2 Preview model and a new TOO_MANY_TOOL_CALLS FinishReason value. HttpOptions.timeout now acts as a per-attempt deadline, and the Interaction resource gains a top-level errors array.

Adds a new environment resource, Google Maps tool grounding with places and routing types, and per-request retryOptions in the API client.

GenerateContentConfig now supports audio_transcription_config and audio transcription in parts, with a flat language_codes field on AudioTranscriptionConfig. Deprecation warnings added for Imagen generate_images, edit_images, and generate_videos using prompt/text/image args, and for LiveConnectConfig.GenerationConfig — all to be removed in the next major version.

Adds a model selector, a custom_vocabulary field for the Live API, a queued status for interaction requests, and makes ASR fields public in v1main and Interactions.

Adds the Triggers resource for workflow automation, a history configuration API, and AntigravityAgentConfig for headless object detection. Also includes per-modality prompt token counts in gemini-embedding-2 embedding responses and support for preserving custom headers in GAOS.

Network config now includes environment_id, and GenerationConfig supports response_format and translation_config. Added Tool.exa_ai_search for Gemini Enterprise API, retrieval result steps, and support for function declarations during agent creation. Fixed interactions to accept both dict and list[dict] for transform, and removed cached_content, presence_penalty, and frequency_penalty from interactions while exposing safety_settings and labels.

Adds video generation and response format parameters, ComputerUse.disabled_safety_policies to both Interactions and Models, and usage fields for steps.