Docs / AI Staging / Providers

AI Staging Providers

BetterStage AI Staging ships with 25 provider presets plus a free-form Custom option, so you're never locked to one vendor. Nearly every modern model speaks the OpenAI chat protocol, so a preset just pre-fills the base URL and a sensible default model — pick one in Settings > AI Staging, paste your key, and go. Prefer zero setup? BetterStage AI runs a managed model for you with no key at all.

Managed — no key needed

BetterStage AI — an optional subscription that runs AI Staging on BetterStage's servers. No API key, no base URL, no model to choose. It's the default, zero-config option; everything below is for bringing your own provider on any Pro license.

Hosted providers

The big labs and routers. They all speak the OpenAI chat protocol — pick a preset, paste a key, go. Some run separate China / International endpoints, so BetterStage gives you a region toggle where the keys differ.

ProviderDefault modelBase URLGet a key
OpenAIgpt-4.1-miniplatform.openai.com
Claude (Anthropic)Uses Anthropic's native message format (x-api-key).claude-sonnet-4-6console.anthropic.com
Google Geminigemini-2.5-flashaistudio.google.com
OpenRouterOne key, hundreds of models from every lab.openai/gpt-4.1-miniopenrouter.ai
Groqllama-3.3-70b-versatileconsole.groq.com
Together AImeta-llama/Llama-3.3-70B-Instruct-Turbotogether.ai
Mistralmistral-large-latestconsole.mistral.ai
xAI (Grok)grok-4console.x.ai
DeepSeekdeepseek-chatplatform.deepseek.com
Moonshot (Kimi)Region toggle: .cn / .aimoonshot-v1-128kplatform.moonshot.cn
Zhipu (GLM)Region toggle: bigmodel.cn / z.aiglm-4.6open.bigmodel.cn
Alibaba QwenRegion toggle: dashscope / dashscope-intlqwen-plusbailian.console.aliyun.com
MiniMaxRegion toggle: .com / .ioMiniMax-Text-01platform.minimaxi.com
Tencent Hunyuanhunyuan-turbos-latestconsole.cloud.tencent.com
ByteDance Doubao (Volcengine)Needs a per-deployment endpoint ID rather than a model name.Endpoint ID (ep-…)console.volcengine.com
Xiaomi MiMomimo-v2.5-proplatform.xiaomimimo.com
StepFunstep-2-16kplatform.stepfun.com

Aggregators & inference clouds

Open-weight model hosts. Browse their catalog and paste the model id you want.

ProviderDefault modelBase URLGet a key
SiliconFlowRegion toggle: .cn / .comdeepseek-ai/DeepSeek-V3cloud.siliconflow.cn
Novita AIbrowse catalognovita.ai
NVIDIA NIMbrowse catalogbuild.nvidia.com
Hugging FaceQwen/Qwen3-Next-80B-A3B-Instructhuggingface.co

Enterprise

Bring your own cloud deployment.

ProviderDefault modelBase URLGet a key
Azure OpenAIType your deployment name instead of a model id.Deployment nameportal.azure.com

Local & self-hosted

Run entirely on your own machine — no key, no data leaving your Mac.

ProviderDefault modelBase URLGet a key
LM Studio (local)your loaded modelNot required (localhost)
Ollama (local)llama3.1Not required (localhost)
Ollama Clouddeepseek-v3.2ollama.com
Custom (OpenAI-compatible)Any OpenAI-compatible proxy or self-hosted server.anyyour endpointyour key

Good to know

  • Any OpenAI-compatible endpoint works — if your provider isn't listed, use the Custom preset with its base URL and model name.
  • Live model lists — most presets fetch the models your key can actually use; the rest fall back to a curated list.
  • Per-provider keys — each preset remembers its own key, so switching providers never loses one.
  • Region toggle — providers with split China / International endpoints (Qwen, Moonshot, Zhipu, MiniMax, SiliconFlow) let you pick the right base URL.
  • Test Connection — sends a real chat ping so you know the endpoint, key, and model all work before you rely on them.

See AI Staging for how to write a Recipe and what data is sent to your provider.