Advanced CLI Arguments¶
Advanced arguments control backend selection, ports, host binding, and environment files. Most users only need scripts/start_unified.sh; use these options when you connect remote inference, run multiple instances, or expose WebUI on a server.
When to Use Advanced Arguments¶
Use advanced arguments when:
- Default ports are already occupied.
- WebUI needs to be reachable from another machine on the LAN.
- A model should use Mock, local, OmniRT, or direct WebSocket backend.
- Model paths, provider keys, mirrors, and endpoints should be managed in an env file.
Backend Selection Arguments¶
--mock¶
The lightest startup mode:
It uses the built-in Mock backend and does not load model weights or require GPU.
--backend¶
Select the backend type for a model:
Accepted values:
mock: no real model, useful for flow validation.local: load the model inside OpenTalking.omnirt: call an independent OmniRT inference service.direct_ws: connect through a direct WebSocket backend.
--model¶
Specify which model receives the backend override:
The script sets the default model and an environment variable such as OPENTALKING_QUICKTALK_BACKEND=local.
Service Port Arguments¶
--api-port¶
Set the OpenTalking API / unified backend port:
--web-port¶
Set the WebUI dev server port:
When you change both API and WebUI ports, pass both arguments together.
--host¶
Set the WebUI bind host:
Use 0.0.0.0 only when WebUI should be reachable from outside the local machine, and check firewall or security group rules.
Remote Inference Arguments¶
--omnirt¶
Set the OmniRT service URL when using --backend omnirt:
bash scripts/start_unified.sh \
--backend omnirt \
--model flashtalk \
--omnirt http://127.0.0.1:9000
If --omnirt is omitted, set OMNIRT_ENDPOINT first.
--env¶
Load a quickstart environment file:
cp scripts/quickstart/env.example .env.quickstart
bash scripts/start_unified.sh --env .env.quickstart --backend local --model quicktalk
Use env files for model paths, TTS provider keys, mirrors, default ports, and endpoints. Do not commit files that contain secrets.
Common Combinations¶
Mock Local Validation¶
Best for first-time installation, WebUI checks, and basic interaction flow.
QuickTalk Local Model¶
Use this on a local GPU machine after model weights and dependencies are ready.
OmniRT Remote Model¶
bash scripts/start_unified.sh \
--backend omnirt \
--model quicktalk \
--omnirt http://127.0.0.1:9000
Use this when inference runs outside OpenTalking and OpenTalking handles WebUI, sessions, and orchestration.
Custom Ports¶
Useful when debugging multiple branches or configurations on one machine.
Common Issues¶
--backend omnirt Requires an Endpoint¶
Pass --omnirt, or set:
WebUI Opens but Session Creation Fails¶
Check API logs and model backend health. For OmniRT, verify that the endpoint is reachable and that the selected model is supported by the remote backend.
WebUI Cannot Be Reached from LAN¶
Start with --host 0.0.0.0, then check firewall, security group, and WebUI port access.