China Providers
Notemd supports China-region and China-origin providers through mostly OpenAI-compatible endpoints. The current set includes DeepSeek, Qwen, Qwen Code, Doubao, Moonshot, Xiaomi MiMo, GLM, MiniMax, Baidu Qianfan, SiliconFlow, Huawei Cloud MaaS, and Z AI. The value is regional reach and cost control; the risk is provider-specific model naming and account gating.
This is part of the Obsidian AI Knowledge Management Guide.
Setup
Create the provider account, generate an API key, then pick the matching Notemd provider preset instead of using a generic OpenAI profile. The preset owns the default Base URL, model, API test mode, model-discovery behavior, and any provider-specific UI fields.
Use per-task model selection aggressively here. For example, keep DeepSeek or GLM on reasoning-heavy workflows, use Qwen or Moonshot for broad Chinese-language work, and route coding tasks to Qwen Code when that account exposes the coder models.
Endpoint And Authentication
| Provider | Default Model | Base URL |
|---|---|---|
| DeepSeek | deepseek-v4-pro | api.deepseek.com |
| Qwen | qwen3-235b-a22b | dashscope.aliyuncs.com/compatible-mode/v1 |
| Qwen Code | qwen3-coder-plus | dashscope.aliyuncs.com/compatible-mode/v1 |
| Doubao | (endpoint ID required) | ark.cn-beijing.volces.com/api/v3 |
| Moonshot | kimi-k2-0905-preview | api.moonshot.cn/v1 |
| Xiaomi MiMo | mimo-v2.5-pro | api.xiaomimimo.com/v1 |
| GLM | glm-5 | open.bigmodel.cn/api/paas/v4 |
| MiniMax | MiniMax-M2.7 | api.minimaxi.com/v1 |
| Baidu Qianfan | ernie-4.5-turbo-32k | qianfan.baidubce.com/v2 |
| SiliconFlow | Qwen/QwQ-32B | api.siliconflow.cn/v1 |
| Huawei Cloud MaaS | DeepSeek-V3 | api.modelarts-maas.com/v1 |
| Z AI | glm-5 | api.z.ai/api/paas/v4 |
Note: Doubao requires an Ark endpoint ID (format: ep-xxxxxxxx-xxxx-xxxx) instead of a standard model name.
Most of these providers use the shared OpenAI-compatible runtime, but their account and model semantics differ. Keep the preset-specific Base URL unless you have verified the account's alternate endpoint.
Model Discovery
DeepSeek, Qwen, Qwen Code, Doubao, Moonshot, Xiaomi MiMo, GLM, Z AI, MiniMax, Baidu Qianfan, and SiliconFlow use OpenAI-compatible model discovery when the provider supports listing. Huawei Cloud MaaS uses the huaweicloud-modelarts-models mode because ModelArts deployment metadata is not identical to a generic /v1/models list.
Do not treat discovery failure as proof that chat is impossible. Several providers allow chat while limiting model-list permissions. In that case, keep an explicit model ID in the profile and validate with the connection test.
Troubleshooting
- Auth succeeds but chat fails: the model name may be outside the account's entitlement or region.
- Doubao fails with a normal model name: use the Ark endpoint ID in the model field when your deployment requires it.
- Reasoning responses are truncated: check provider-specific max output tokens and Notemd's model-aware max-token settings.
- Regional networking varies: use the provider's China-region endpoint when the vault machine is inside China; use the international endpoint only when the account and network path are both valid.
When To Use
Use China providers when regional latency, payment/account access, Chinese-language quality, or cost is the primary constraint. Prefer direct OpenAI, Anthropic, Google, or local providers when you need a simpler protocol boundary or stronger cross-provider debugging consistency.
Next Steps
- LLM Providers -- Full configuration, per-task models, and retry logic
- Troubleshooting -- Fix regional connectivity and authentication errors