Changelog

Scribe v2

We launched Scribe v2, the new state of the art transcription model. Learn more about Scribe v2 in the docs.

Agents Platform

• Timezone support for batch call scheduling: You can now select a timezone when scheduling batch calls. The scheduled time is converted to UTC based on your chosen timezone. Your browser’s timezone is automatically selected by default, and validation prevents scheduling calls for times that have already passed in the selected timezone. This makes it easier to schedule outbound calls for the right time in your recipients’ time zones.
• Knowledge Base source file URL: Added a new endpoint to retrieve the original source file URL for knowledge base documents, enabling direct access to uploaded files.
• LLM fallback cascade timeout: Added cascade_timeout_seconds configuration option for agent backup LLM configs, allowing control over how long to wait before cascading to the next LLM. Default is 8 seconds with an allowed range of 2-15 seconds.
• Soft timeout LLM-generated messages: Added use_llm_generated_message option to soft timeout configuration. When enabled, the agent will generate a contextual message using the LLM instead of using a predefined message when soft timeout triggers.
• Knowledge Base folder navigation: Knowledge Base document responses now include folder_path field showing the path segments from root to parent folder, making it easier to understand document hierarchy.
• Conversation filtering by initiation source: The Get Conversations endpoint now supports filtering by conversation_initiation_source query parameter.

Dubbing

• New transcript format endpoint: Added GET /v1/dubbing/{dubbing_id}/transcripts/{language_code}/format/{format_type} endpoint supporting srt, webvtt, and json output formats. The previous GET /v1/dubbing/{dubbing_id}/transcript/{language_code} endpoint is now deprecated.
• Required filename for dubbed files: Knowledge Base file models now require a filename field.

Speech to Text

• Entity detection: Added entity_detection option to Speech-to-Text requests. Accepts 'all', specific entity type strings, or an array of entity types. Detected entities are returned in a new entities response field using the DetectedEntity schema.
• Keyterm prompting: Added keyterms array parameter to Speech-to-Text requests for biasing transcription toward specific terms or phrases.

SDK Releases

Python SDK

• v2.29.0 - Added entity detection and keyterm prompting for Speech-to-Text, LLM cascade timeout configuration, soft timeout LLM message generation, and batch call timezone support
• v2.28.0 - Added agent versioning fields, voice collection IDs, batch call enhancements, and phone number labels

JavaScript SDK

• v2.30.0 - Added entity detection and keyterm prompting for Speech-to-Text, LLM cascade timeout configuration, soft timeout LLM message generation, and batch call timezone support
• v2.29.0 - Added agent versioning fields, voice collection IDs, batch call enhancements, and phone number labels

MCP Server

• v0.9.1 - Added Gemini Extension support, fixed path handling for non-absolute output_directory and missing base_path

API