Product line · Models
Overseas models reach teams in China, Chinese models reach teams abroad. One key, one interface, one usage bill.
Model access
Both directions · sample status
Overseas models · called from China
Chinese models · called from abroad
The model list in the console is authoritative. New models are added there first.
Routing
The caller names a model. The gateway decides the route from the caller's origin and region.
Teams in China and outbound businesses call GPT, Claude, Gemini and others with a request body identical to the official API.
Overseas customers call DeepSeek, Qwen, Kimi and others, entering through the route nearest to the caller.
Capabilities
A common interface is the starting point. Usage, routing, logs and delivery decide whether it holds up over time.
OpenAI-compatible. Change base_url and key to move to another model, without touching application code.
Overseas models into China and Chinese models out, sharing one account and one billing model.
Token usage by project, member and model, with monthly budgets and per-call caps.
Multiple routes with fallback when one degrades. Timeouts and retry policy are configurable.
Time, model, token count and caller recorded. Retention scope is set in the contract.
Shared access works out of the box, or the gateway runs inside your own network.
Integration
Existing OpenAI SDK code usually stays as it is.
After signing or self-registration, issue a key per project in the console.
Set base_url to the gateway and pick a model id from the console list.
Token usage and cost by project, model and member.
curl https://api.walmind.ai/v1/chat/completions \
-H "Authorization: Bearer $WALMIND_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "MODEL_ID",
"messages": [{"role": "user", "content": "hello"}]
}'MODEL_ID comes from the console model list. For the OpenAI SDK, only base_url and api_key change.
Typical use
Copy, support replies and product descriptions generated per target language, with the model chosen for that language.
The same code path switched between models, comparing quality and cost on real traffic.
DeepSeek, Qwen and others reached from local networks, settled on one bill.
When the primary model is unavailable, switch to the fallback by changing the model name.
Onboarding
Companies sign a contract. Individuals and small teams can register and top up.
Qualification review, signed contract, bank transfer, monthly invoicing.
Register with an email, top up, generate a key, pay as you go.
FAQ
Overseas: OpenAI, Anthropic, Google, xAI, Mistral. Chinese: DeepSeek, Qwen, Kimi, Zhipu, Doubao, MiniMax. The console list is authoritative.
It is OpenAI-compatible. Change base_url and api_key; the request body stays the same.
Retention is set in the contract. It can be limited to time, model, token count and caller.
Per token, with different rates per model. Budgets and per-call caps are set in the console.
Contract customers are invoiced monthly.
Yes. The gateway runs on your premises; model egress still goes through WALMIND routes.
Next step
Tell us the use case, the target model and the daily call volume. We reply with an integration plan and pricing.
Replies within one business day, 9:00–18:00 GMT+8.