arcane_voice_proxy 1.3.1 copy "arcane_voice_proxy: ^1.3.1" to clipboard
arcane_voice_proxy: ^1.3.1 copied to clipboard

Published 7 days ago • verified publisherarcane.art
SDKFlutter
PlatformAndroidiOSLinuxmacOSWindows
→

Metadata

Server-side realtime voice proxy for OpenAI, Gemini, Grok, ElevenLabs, and Twilio voice calls.

More...

1.3.1 #

  • Fixed Twilio Media Streams barge-in behavior by clearing Twilio's buffered assistant audio when caller speech starts during an active response.
  • Fixed Grok provider-VAD barge-in so provider-detected caller speech cancels the active response before Twilio playback is cleared.

1.3.0 #

  • Added Twilio Voice inbound-call support through a built-in voice webhook and bidirectional Media Streams websocket bridge.
  • Added ArcaneVoiceTwilioConfig, ArcaneVoiceTwilioGateway, ArcaneVoiceTwilioCallContext, and Twilio audio/TwiML helpers to the public proxy API.
  • Attached Twilio From, To, CallSid, AccountSid, direction, status, and custom parameters to sessionContextJson for server-side session resolution.
  • Added typed Twilio call-context parsing helpers so hosts can map caller phone numbers to per-session prompts, RAG scopes, and proxy tool registries.
  • Documented Twilio setup, TLS/WSS requirements, and phone-number scoped session resolver patterns.

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.

1.1.3 #

  • 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.

1.1.2 #

  • 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.

1.1.1 #

  • 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.

1.1.0 #

  • 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.

1.0.0 #

  • Initial version.

← Metadata

0
likes
140
points
389
downloads

Documentation

API reference

Publisher

verified publisherarcane.art

Weekly Downloads

Metadata

Server-side realtime voice proxy for OpenAI, Gemini, Grok, ElevenLabs, and Twilio voice calls.

Repository (GitHub)

License

GPL-3.0 (license)

Dependencies

arcane_voice_models, fast_log, web_socket_channel

More

Packages that depend on arcane_voice_proxy

Back