Configuration and deployment

Runtime configuration is centralized in app.AppConfig. Use app.DefaultConfigFromEnv() to load environment variables, then override fields in Go when needed.

package main

import "github.com/cavos-io/rtp-agent/app"

func buildApp() (*app.App, error) {
	cfg := app.DefaultConfigFromEnv()
	cfg.Instructions = "You are a concise support agent."
	cfg.LLMProvider = "openai"
	cfg.LLMModel = "gpt-4.1-mini"

	rtpApp, err := app.Init(cfg)
	if err != nil {
		return nil, err
	}

	return rtpApp, nil
}

Core environment variables

Common variables include:

• LIVEKIT_URL
• LIVEKIT_API_KEY
• LIVEKIT_API_SECRET
• RTP_AGENT_INSTRUCTIONS
• RTP_AGENT_LLM_PROVIDER
• RTP_AGENT_LLM_MODEL
• RTP_AGENT_STT_PROVIDER
• RTP_AGENT_STT_MODEL
• RTP_AGENT_VAD_PROVIDER
• RTP_AGENT_TTS_PROVIDER
• RTP_AGENT_TTS_MODEL
• RTP_AGENT_TTS_VOICE
• RTP_AGENT_REALTIME_PROVIDER
• RTP_AGENT_REALTIME_MODEL
• RTP_AGENT_AVATAR_PROVIDER
• RTP_AGENT_TURN_DETECTOR_PROVIDER
• RTP_AGENT_SMART_TURN_GRPC_ADDR (gRPC inference endpoint when provider is cavos; defaults to localhost:9001)

Provider API keys are provider-specific, for example OPENAI_API_KEY, DEEPGRAM_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY, and LIVEKIT_API_KEY.

Deployment shape

Build and run a Go binary that initializes app.App and passes its server to cli.RunApp. Worker transport, credentials, and provider selection should come from environment or deployment configuration.

Do not document a deployment target as supported unless the repository contains the corresponding server, transport, or integration code.