China-region providers

Revka has first-party support for the major China-based model vendors — Zhipu GLM, Z.AI, MiniMax, Alibaba Qwen/DashScope, Moonshot/Kimi, Baidu Qianfan, ByteDance Doubao, and Alibaba Bailian. Almost all of them ship two regional deployments: a CN endpoint (mainland data centers, typically lower latency from inside China) and a Global/International endpoint. Revka picks the right base URL automatically from the provider alias you choose, so switching regions is a one-word config change rather than a URL edit.

Use this page when you want to run any of these models, when you need to route a deployment to the CN versus the Global endpoint, or when you want to reuse a Qwen Code or MiniMax subscription token instead of an API key. For the full provider table see the provider catalog; for fallbacks and hint-based routing see routing, reliability & tuning.

How region aliases work

Most China providers expose paired aliases like qwen / qwen-intl, glm / glm-cn, or minimax / minimax-cn. The alias you set as default_provider decides which base URL Revka calls — you never set the endpoint by hand. The CN vs Global aliases table at the bottom lists every alias.

Quick start

Every China provider follows the same three-step pattern: set the provider alias, set its API key (config or env var), pick a model.

1. Pick a provider alias. For example qwen (CN) or qwen-intl (Global).

2. Set the key. Either run onboarding non-interactively or set the provider’s env var:

   revka onboard --provider qwen --api-key "YOUR_DASHSCOPE_KEY"
   # or
   export DASHSCOPE_API_KEY="YOUR_DASHSCOPE_KEY"

3. Choose a model and chat.

   ~/.revka/config.toml

   default_provider = "qwen"
   default_model    = "qwen-max"

   echo "Hello" | revka agent

Credential resolution order

For every provider, Revka resolves the key in this order: the explicit api_key in config.toml → the provider-specific env var(s) listed below → the generic fallbacks REVKA_API_KEY then API_KEY. The provider-specific env var is the recommended path so a stray generic key never bleeds into the wrong provider.

Zhipu GLM

The glm provider integrates Zhipu AI’s GLM models through an OpenAI-compatible surface. GLM API keys use Zhipu’s distinctive id.secret format (for example abc123.xyz789). The provider runs with the responses-API fallback disabled, so it always speaks /chat/completions and never probes /responses.

	CN	Global
Base URL	https://open.bigmodel.cn/api/paas/v4	https://api.z.ai/api/paas/v4
Aliases	glm-cn, zhipu-cn, bigmodel	glm, zhipu, glm-global, zhipu-global
Auth	Bearer (id.secret)	Bearer (id.secret)
Env var	GLM_API_KEY	GLM_API_KEY

# ~/.revka/config.toml — GLM Global
default_provider = "glm"
default_model    = "glm-4.7"
api_key          = "<id>.<secret>"

# GLM China (open.bigmodel.cn)
default_provider = "glm-cn"
default_model    = "glm-4.7"
api_key          = "<id>.<secret>"

id.secret key format

A GLM key that is missing the dot (.) separator will fail authentication. Copy the whole id.secret string from the Zhipu console with no surrounding whitespace.

Z.AI

Z.AI exposes Zhipu’s coding-plan deployment on a separate path (/api/coding/paas/v4) from the standard GLM endpoint. Pick zai when you have a Z.AI coding plan subscription; pick glm for the standard GLM API.

	CN	Global
Base URL	https://open.bigmodel.cn/api/coding/paas/v4	https://api.z.ai/api/coding/paas/v4
Aliases	zai-cn, z.ai-cn	zai, z.ai, zai-global, z.ai-global
Auth	Bearer (id.secret)	Bearer (id.secret)
Env var	ZAI_API_KEY	ZAI_API_KEY

# ~/.revka/config.toml — Z.AI Global
default_provider = "zai"
default_model    = "glm-4.7"
api_key          = "<id>.<secret>"

Z.AI / GLM provider setup

Z.AI and GLM share the same id.secret key format but use different endpoints and env vars. Use this checklist to wire either one:

1. Get a key. Sign up at the Z.AI or Zhipu (bigmodel.cn) console and generate a key. The format is always id.secret.

2. Choose the right alias. Use zai/zai-cn for a Z.AI coding plan, or glm/glm-cn for the standard GLM API. Append -cn for the mainland endpoint.

3. Set the matching env var. ZAI_API_KEY for Z.AI, GLM_API_KEY for GLM. Each alias family only reads its own env var.

4. Verify with curl before pointing Revka at it:

   curl -X POST "https://api.z.ai/api/coding/paas/v4/chat/completions" \
     -H "Authorization: Bearer YOUR_ZAI_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{"model":"glm-4.7","messages":[{"role":"user","content":"Hello"}]}'

5. Confirm Revka sees it:

   revka status
   echo "Hello" | revka agent

Model availability varies

The exact model IDs offered (for example the glm-4.x line and lower-latency air variants) depend on your account and region. When a model ID is rejected, list what your account can see with revka models list --provider zai or query the endpoint’s /models route directly.

MiniMax

MiniMax offers both a Global and a CN deployment. Because some MiniMax models reject a role: system message, Revka merges the system prompt into the first user turn for every MiniMax alias. Keep this in mind for channel conversations whose system prompt carries delivery instructions — that guidance is folded into the user message rather than sent separately.

	CN	Global
Base URL	https://api.minimaxi.com/v1	https://api.minimax.io/v1
Aliases	minimax-cn, minimaxi, minimax-oauth-cn, minimax-portal-cn	minimax, minimax-intl, minimax-io, minimax-global, minimax-portal
Auth	Bearer	Bearer
Env var	MINIMAX_OAUTH_TOKEN, then MINIMAX_API_KEY	MINIMAX_OAUTH_TOKEN, then MINIMAX_API_KEY

# ~/.revka/config.toml — MiniMax Global
default_provider = "minimax"
default_model    = "MiniMax-Text-01"
api_key          = "YOUR_MINIMAX_API_KEY"

MiniMax region routing

MiniMax aliases self-select their endpoint: any alias matched as international routes to api.minimax.io, and any alias matched as CN routes to api.minimaxi.com. Use minimax or minimax-intl for Global, and minimax-cn or minimaxi for China — no api_url needed.

MiniMax OAuth (subscription token)

Setting api_key = "minimax-oauth" is a placeholder that triggers automatic OAuth token resolution instead of treating the literal string as a key. Revka then checks, in order: MINIMAX_OAUTH_TOKEN, MINIMAX_API_KEY, and finally the token-refresh endpoint using MINIMAX_OAUTH_REFRESH_TOKEN. Region for the refresh call is taken from the alias or from MINIMAX_OAUTH_REGION.

~/.revka/config.toml

default_provider = "minimax-oauth"
api_key          = "minimax-oauth"   # placeholder, not a literal key

export MINIMAX_OAUTH_TOKEN="..."          # preferred direct token
# or, to auto-refresh:
export MINIMAX_OAUTH_REFRESH_TOKEN="..."
export MINIMAX_OAUTH_REGION="global"      # or "cn"

Variable	Meaning
MINIMAX_OAUTH_TOKEN	Direct OAuth access token (checked first)
MINIMAX_API_KEY	Standard API key fallback
MINIMAX_OAUTH_REFRESH_TOKEN	Enables automatic token refresh
MINIMAX_OAUTH_REGION	global or cn — selects the refresh endpoint
MINIMAX_OAUTH_CLIENT_ID	Override the default OAuth client ID

The refresh endpoints are https://api.minimax.io/oauth/token (Global) and https://api.minimaxi.com/oauth/token (CN). See OAuth & subscription providers for the broader OAuth model.

Qwen / DashScope

Alibaba’s Qwen models are served through DashScope’s OpenAI-compatible endpoint. There are three separate regional deployments — CN, International, and US — plus an OAuth variant for Qwen Code CLI subscribers. Vision (base64 images) is enabled for Qwen.

Region	Base URL	Aliases
CN	https://dashscope.aliyuncs.com/compatible-mode/v1	qwen, dashscope, qwen-cn, dashscope-cn
International	https://dashscope-intl.aliyuncs.com/compatible-mode/v1	qwen-intl, dashscope-intl, qwen-international, dashscope-international
US	https://dashscope-us.aliyuncs.com/compatible-mode/v1	qwen-us, dashscope-us

All three direct-key variants authenticate with DASHSCOPE_API_KEY (Bearer).

# ~/.revka/config.toml — Qwen International
default_provider = "qwen-intl"
default_model    = "qwen-max"

export DASHSCOPE_API_KEY="YOUR_DASHSCOPE_KEY"

Qwen Code OAuth

The qwen-code (alias qwen-oauth) variant reuses a Qwen Code CLI subscription instead of a direct key, analogous to other subscription providers. Set the placeholder key and let Revka resolve the token:

~/.revka/config.toml

default_provider = "qwen-code"
api_key          = "qwen-oauth"   # placeholder

Variable	Meaning
QWEN_OAUTH_TOKEN	Explicit OAuth access token
QWEN_OAUTH_REFRESH_TOKEN	Enables automatic refresh when the cached token expires
QWEN_OAUTH_RESOURCE_URL	Endpoint override returned by the OAuth session
QWEN_OAUTH_CLIENT_ID	Override the default OAuth client ID

Revka also auto-discovers credentials written by the Qwen Code CLI at ~/.qwen/oauth_creds.json. When no resource URL is supplied, the OAuth variant falls back to the CN base URL. See OAuth & subscription providers for the full flow.

Moonshot / Kimi

Moonshot’s Kimi models ship CN and International deployments. Note the alias mapping is asymmetric: the bare moonshot and kimi aliases point at the CN endpoint; use the -intl/-global aliases for the international endpoint.

	CN	International
Base URL	https://api.moonshot.cn/v1	https://api.moonshot.ai/v1
Aliases	moonshot, kimi, moonshot-cn, kimi-cn	moonshot-intl, moonshot-global, kimi-intl, kimi-global
Auth	Bearer	Bearer
Env var	MOONSHOT_API_KEY	MOONSHOT_API_KEY

# ~/.revka/config.toml — Kimi International
default_provider = "kimi-intl"
default_model    = "kimi-k2"
api_key          = "YOUR_MOONSHOT_API_KEY"

Kimi for Coding

The separate kimi-code provider (aliases kimi_coding, kimi_for_coding) targets Moonshot’s coding endpoint at https://api.kimi.com/coding/v1 and reads KIMI_CODE_API_KEY (falling back to MOONSHOT_API_KEY). It is distinct from the chat-oriented moonshot/kimi aliases above.

Qianfan / Baidu

Baidu’s Qianfan platform (formerly Wenxin/ERNIE) is served from a single endpoint.

Field	Value
Base URL	https://qianfan.baidubce.com/v2
Aliases	qianfan, baidu
Auth	Bearer
Env var	QIANFAN_API_KEY

~/.revka/config.toml

default_provider = "qianfan"
default_model    = "ernie-4.5-turbo"
api_key          = "YOUR_QIANFAN_API_KEY"

You can override the base URL with api_url if Baidu routes your account to a different host.

Doubao / Volcengine

ByteDance’s Doubao models run on the Volcengine Ark platform. The endpoint is mainland-hosted.

Field	Value
Base URL	https://ark.cn-beijing.volces.com/api/v3
Aliases	doubao, volcengine, ark, doubao-cn
Auth	Bearer
Env vars	ARK_API_KEY, then VOLCENGINE_API_KEY, then DOUBAO_API_KEY

~/.revka/config.toml

default_provider = "doubao"
default_model    = "doubao-pro-32k"
api_key          = "YOUR_ARK_API_KEY"

Ark endpoint IDs

Volcengine Ark frequently addresses a model by a per-account endpoint ID (e.g. ep-2024...) rather than a public model name. Use whatever identifier the Ark console gives you as default_model.

Bailian

Alibaba Cloud Bailian (Model Studio) is served from a dedicated DashScope coding host. Vision is enabled for this provider.

Field	Value
Base URL	https://coding.dashscope.aliyuncs.com/v1
Aliases	bailian, aliyun-bailian, aliyun
Auth	Bearer
Env vars	BAILIAN_API_KEY, then DASHSCOPE_API_KEY

~/.revka/config.toml

default_provider = "bailian"
default_model    = "qwen-max"
api_key          = "YOUR_BAILIAN_API_KEY"

Because Bailian falls back to DASHSCOPE_API_KEY, a single DashScope key can serve both the qwen and bailian aliases if your account is enabled for both.

CN vs Global aliases

Every China provider’s region is selected purely by the alias you set as default_provider. This table is the authoritative map of which alias resolves to which endpoint.

Provider	CN endpoint	CN aliases	Global / Intl endpoint	Global / Intl aliases
GLM	https://open.bigmodel.cn/api/paas/v4	glm-cn, zhipu-cn, bigmodel	https://api.z.ai/api/paas/v4	glm, zhipu, glm-global, zhipu-global
Z.AI	https://open.bigmodel.cn/api/coding/paas/v4	zai-cn, z.ai-cn	https://api.z.ai/api/coding/paas/v4	zai, z.ai, zai-global, z.ai-global
MiniMax	https://api.minimaxi.com/v1	minimax-cn, minimaxi, minimax-oauth-cn, minimax-portal-cn	https://api.minimax.io/v1	minimax, minimax-intl, minimax-io, minimax-global, minimax-portal
Qwen / DashScope	https://dashscope.aliyuncs.com/compatible-mode/v1	qwen, dashscope, qwen-cn, dashscope-cn	Intl: https://dashscope-intl.aliyuncs.com/compatible-mode/v1 · US: https://dashscope-us.aliyuncs.com/compatible-mode/v1	qwen-intl, dashscope-intl, qwen-international · qwen-us, dashscope-us
Moonshot / Kimi	https://api.moonshot.cn/v1	moonshot, kimi, moonshot-cn, kimi-cn	https://api.moonshot.ai/v1	moonshot-intl, moonshot-global, kimi-intl, kimi-global
Qianfan / Baidu	https://qianfan.baidubce.com/v2	qianfan, baidu	(single endpoint)	—
Doubao / Volcengine	https://ark.cn-beijing.volces.com/api/v3	doubao, volcengine, ark, doubao-cn	(single endpoint)	—
Bailian	https://coding.dashscope.aliyuncs.com/v1	bailian, aliyun-bailian, aliyun	(single endpoint)	—

Watch the asymmetric Moonshot aliases

For most providers the bare alias is the Global endpoint, but for Moonshot/Kimi the bare moonshot and kimi aliases resolve to the CN endpoint. Use kimi-intl / moonshot-intl if you mean the international deployment.

Overriding the endpoint

The region aliases cover the standard deployments, but you can always pin an explicit base URL with api_url. This is useful for a private gateway, a regional mirror, or a host Alibaba/Baidu routed your account to:

default_provider = "qianfan"
api_url          = "https://your-qianfan-mirror.example.com/v2"
api_key          = "YOUR_QIANFAN_API_KEY"

If no named alias fits at all, fall back to the generic custom: prefix described in local, self-hosted & custom endpoints.

Related pages

Provider quickstart Pick a provider, set the key, run your first chat.

Provider catalog The full table of every provider, alias, and env var.

OAuth & subscription providers MiniMax and Qwen Code OAuth token flows in depth.

Routing, reliability & tuning Fallback chains, hint routing, timeouts and retries.