arcane_voice_proxy 1.1.4 changelog

1.1.4 #

Added ArcaneVoiceProxyVadMode to proxy setup so hosts can choose auto, local, or provider turn detection behavior.
Defaulted proxy VAD mode to auto, which prefers provider-native VAD when a provider adapter supports it.
Wired OpenAI, Grok, and Gemini sessions to switch between local/manual turn handling and provider-native activity detection from the proxy API.

Fixed ElevenLabs client_tool_result payload serialization so structured tool outputs are returned as JSON strings instead of raw objects.
Prevented ElevenLabs websocket sessions from being closed after successful proxy tool execution due to invalid result message shapes.

Normalized generic JSON Schema into the stricter ElevenLabs client-tool schema subset before tool registration.
Removed unsupported keys like additionalProperties and synthesized valid literal descriptions for array item and leaf parameter schemas so ElevenLabs accepts tools like query_record.

Fixed ElevenLabs client tool registration to send schemas under parameters instead of params, so agents can see and use full tool argument definitions.
Added a regression test covering ElevenLabs client tool schema generation.

Added session resolution hooks for host-owned auth, config overrides, and per-session tools.
Added lifecycle callbacks for session start, usage observation, tool execution, and session stop.
Normalized provider usage reporting across OpenAI, Gemini, Grok, and ElevenLabs flows.
Removed deprecated server* alias API so the public surface uses proxyTools and ArcaneVoice... names only.
Updated docs and examples for authenticated server-hosted sessions.