Models overview

Status: implemented for provider interfaces and many adapters; partial for cloud model catalog behavior.

Use this section to choose which model interface your agent needs.

rtp-agent treats models as Go interfaces plus provider adapters. The app layer wires configured adapters into agent.Agent and agent.AgentSession, but you can also construct providers directly when you need provider-specific options.

Choose by runtime path

• Use llm.LLM for text/chat generation in the speech pipeline.
• Use stt.STT to turn user audio into speech events.
• Use tts.TTS to synthesize agent speech into audio frames.
• Use llm.RealtimeModel for realtime model sessions.
• Use agent.AvatarProvider when a provider drives virtual-avatar state.
• Use VAD providers when endpointing or speech activity should be separated from STT.

Choose providers by source capability

A provider supports a capability only when its adapter package contains the matching source file, such as llm.go, stt.go, tts.go, realtime.go, or avatar.go. Model names and provider option details belong to the adapter source and tests, not a hand-maintained marketing catalog.

Evidence:

• core/llm/llm.go
• core/stt/stt.go
• core/tts/tts.go
• core/vad/vad.go
• core/agent/avatar.go
• adapter/*