OptionalargumentsTool-call arguments. Inworld may emit these either pre-parsed (object) or as a JSON string; the
driver normalizes both to the Core contract's JSON-string Arguments shape.
OptionalaudioA base64-encoded chunk of synthesized output audio (Realtime TTS-2).
Optionalcall_A correlation id for a tool call, echoed back on the tool result.
OptionalcodeProvider error code on an error frame.
OptionalfatalWhether the provider classified an error frame as fatal (transport/credential death).
Optionalinput_Incremental input-token count on a usage frame.
OptionalmessageHuman-readable message on an error frame.
OptionalnameThe tool name on a tool-call frame.
Optionaloutput_Incremental output-token count on a usage frame.
Optionalsession_Provider-assigned session id, surfaced on the session-ready frame.
OptionaltextTranscribed text for a transcript frame (user STT or assistant TTS transcript).
OptionaltypeThe event discriminator (e.g. session.updated, output_audio.delta, response.done).
A parsed inbound Inworld Realtime WebSocket frame.
The Inworld protocol multiplexes server events on a
typediscriminator (mirroring thesession.updaterequest shape it accepts), so a single interface carries every inbound field the driver reads. Fields are optional because which ones are present depends ontype.Remarks
Wire-format binding point — the precise inbound
typestrings and field names are mapped to the documented protocol shape and centralized in InworldRealtimeSession.classifyServerEvent; verify the exact discriminators against a live Inworld endpoint.