v0.13.0

0.13.0 (2026-10-12)

⚠ BREAKING CHANGES

• manage: Details::duration and Details::total_audio are now Option<f64>; Details::channels and Details::streams are now Option<usize>. Update direct field reads to handle None deliberately; an absent legacy meter group does not necessarily represent zero usage.
• manage: Management and Auth::grant calls now use the configured base URL. Code that relied on a custom-base-URL client sending those calls to the hosted API must create a separate default-hosted client with Deepgram::new(api_key) or Deepgram::with_temp_token(token). A base URL with a path prefix must end in / to preserve that prefix; without the trailing slash, RFC 3986 resolution replaces the final path segment.
• Streaming metadata and terminal responses are now non-exhaustive and expose extra metadata; external struct literals must be replaced with deserialized values and exhaustive patterns must include ... Named Whisper and redaction values now parse to their dedicated variants, so code matching CustomId or Other for those strings must match the named variants or compare their wire values instead. Speak::speak_to_file no longer writes a success message to stdout; print that message in the application if needed.
• listen: expose extra metadata on pre-recorded and streaming responses
• listen: StreamResponse::TerminalResponse is non-exhaustive, so pattern matches must include ..; its contained Metadata cannot be constructed externally with a struct literal.

Added

• add Phase 1 quick wins for TTS metadata, Whisper models, redaction, response fields, and captions (65d6c7d)
• agent: add Flux and Aura settings with integer Deepgram Speak V2 expressivity (1e30b0d)
• agent: add listen controls, raw messages, typed server events, and forward-compatible unknown values (1e30b0d)
• agent: preserve typed JSON events and synthesized audio in wire order (1e30b0d)
• captions: add SRT and WebVTT helpers with speaker-safe grouping and millisecond timestamps (65d6c7d)
• listen: expose Entity::raw_value to preserve pre-formatting spoken text (65d6c7d)
• listen: expose extra metadata on pre-recorded and streaming responses (65d6c7d)
• models: add configured-base-URL model listing, including outdated models and expanded metadata (a7187a2)
• read: add TextIntelligence analysis with optional partial-response fields, callbacks, and the standalone read feature (a7187a2)
• self-hosted: add typed distribution credential management with redacted secrets and forward-compatible scopes (49a25b9)
• speak: add streaming TTS options, local validation, bounded close handling, and shared TLS configuration (49a25b9)

Fixed

• agent: prevent inbound audio starvation, bound close handling, require secure remote WebSockets, and redact credentials (1e30b0d)
• listen: fix single-feature cargo docs, terminal transport errors, and response-buffer write stalls (a7187a2)
• listen: keep the streaming worker running when the idle timer fires without keep-alive (#186) (1494612)
• manage: decode usage requests that omit the audio meter group (4622a4a)
• manage: distinguish omitted usage meter groups from zero usage (4622a4a)
• manage: Scopes::get_scope no longer builds its URL with a trailing space (8bea4ed)
• manage: send management and auth requests to the configured base URL (8bea4ed)
• speak: add named Aura-2 models and prevent Flux TTS terminal-error deadlocks under event backpressure (49a25b9)