Ozon AI Call Simulator

Environment configuration

• .env.example: template with all required variables. Copy it to .env.development or .env.production and fill in real values.
• .env.development: used for local development together with docker-compose.dev.yml.
• .env.production: used for production / remote deployment together with docker-compose.yml.

Environment variables reference

• OPENROUTER_API_KEY / OPENAI_API_KEY: API key(s) for your LLM provider (OpenRouter / OpenAI).
• OPENROUTER_MODEL_DEFAULT: default model used by the frontend when calling the LLM.
• FRONTEND_ORIGIN: base URL where the frontend is served (e.g. http://localhost:3000 or your production domain).
• LIVEKIT_URL: WebSocket URL of the LiveKit server used for audio/video sessions.
• LIVEKIT_API_KEY / LIVEKIT_API_SECRET: credentials for authenticating against LiveKit.
• POSTGRES_USER / POSTGRES_PASSWORD / POSTGRES_DB: Postgres credentials and database name.
• TONE_MODELS_DIR: path to the T-One STT models directory used by stt_service.
• DATABASE_URL: full Postgres connection string used by judge, auth, and agent services.
• AUTH_JWT_SECRET: secret key for signing auth JWT tokens (must be strong in production).
• AUTH_TOKEN_TTL_HOURS: lifetime of issued auth tokens in hours.
• LLM_PROVIDER: which LLM backend to use (openrouter or ollama).
• OPENROUTER_BASE_URL: base URL for the OpenRouter API.
• JUDGE_OPENROUTER_MODEL: OpenRouter model used by judge_service.
• AGENT_OPENROUTER_MODEL: OpenRouter model used by agent_service.
• JUDGE_OLLAMA_BASE_URL / JUDGE_OLLAMA_MODEL: Ollama base URL and model used by judge_service.
• AGENT_OLLAMA_BASE_URL / AGENT_OLLAMA_MODEL: Ollama base URL and model used by agent_service.

Using Ollama instead of OpenRouter

• Install Ollama on your machine and pull a model, for example:
  • ollama pull qwen2:7b-instruct-q4_K_M
• Set env vars in .env.development or .env.production:
  • LLM_PROVIDER=ollama
  • OLLAMA_BASE_URL=http://host.docker.internal:11434 (or your Ollama host)
  • OLLAMA_MODEL to the model name you pulled (e.g. qwen2:7b-instruct-q4_K_M)
• Restart Docker services so judge_service and agent_service pick up the new configuration.

LiveKit server configuration (livekit.yaml)

• File locations:
  • livekit/livekit.dev.yaml: used by docker-compose.dev.yml for local development.
  • livekit/livekit.yaml: used by docker-compose.yml for production.
• Ports:
  • port: HTTP/WebSocket signaling port (matches 7880 in docker-compose*).
  • rtc.port_range_start / rtc.port_range_end: UDP port range for media (exposed in docker).
  • rtc.tcp_port (prod only): TCP fallback port for WebRTC (matches 7881 in docker-compose.yml).
• Networking / TURN:
  • rtc.use_external_ip / rtc.node_ip: how the node advertises its IP in ICE candidates.
  • turn.enabled, turn.udp_port, turn.domain: built-in TURN server for clients that cannot use direct UDP.
• Auth keys:
  • keys: API key / secret pairs; must match LIVEKIT_API_KEY / LIVEKIT_API_SECRET in your env files.

How to run for development

Download T-One models from cloud and place them to stt_service/tone_models

https://drive.google.com/drive/folders/1VPnN8XAr2ads2HqjbrIMEPqXgaOvNW4u?usp=sharing

Export environment variables

source .env.development

Run services

docker compose -f docker-compose.dev.yml up -d

Open localhost in browser and enjoy

How to run for production

Export environment variables

source .env.production

Run services

docker compose up -d

Open your production domain in browser