FAQ

Which models are available?

The MVP catalog covers seven providers: Anthropic (Claude 3.5 Sonnet, Claude 3.5 Haiku), OpenAI (GPT-4o, GPT-4o-mini), Google (Gemini 2.0 Flash), DeepSeek (V3, R1), Moonshot (Kimi 128K), Alibaba (Qwen 2.5 72B), and Zhipu (GLM-4.5). The canonical model IDs are listed in the quickstart and returned by GET /v1/models.

Is token-hub really OpenAI-compatible?

Yes. /v1/chat/completions and /v1/embeddings accept the same request bodies and return the same response shapes as OpenAI. Any OpenAI SDK — Python, Node, Go, Ruby, C# — works by overriding base_url to https://api.sandboxclaw.com/v1 and swapping the key.

Where am I billed, and in what currency?

Billing runs through Stripe. You can pay in USD or CNY; switch the currency toggle on /topup. Credits post to your account within ~15 seconds of checkout and stay valid for 12 months. There is no subscription and no monthly minimum. Alipay and WeChat Pay are on the roadmap.

Do I need ICP filing or a mainland China phone number?

No. The service is operated outside mainland China, which is also why developers in China can call Claude and GPT through it without a VPN. The only signup inputs are email (or Google) and a Stripe-accepted payment method.

Are my prompts used to train models?

No. token-hub does not retain prompt or completion content for training, analytics, or resale. Operational metadata — status codes, token counts, latency, model ID — is kept 7 days for debugging and abuse detection, then deleted. Upstream providers have their own retention policies; consult their terms for details.

What are the default rate limits?

60 requests per minute per key and 600 requests per minute per account. Burst above that returns 429 rate_limit_exceeded with a Retry-After header in seconds. Higher limits are granted on request — email support@sandboxclaw.com with your use case and expected RPM.

What happens when an upstream provider is down?

We return 502 upstream_error or 504 upstream_timeout with the provider name in the error body. The request is not billed. Our recommendation is a 3-attempt exponential backoff (1s, 2s, 4s), and if a family repeatedly fails, fall back to a sibling model — for example claude-3-5-haiku for claude-3-5-sonnet, or gpt-4o-mini for gpt-4o. token-hub does not do automatic cross-provider failover; that’s an application decision you own.

Can I get a refund?

Yes. Unused credits are refundable within 7 days of the original topup — email support@sandboxclaw.com with the topup ID. After 7 days, credits stay on your account and expire 12 months from purchase.

Can I revoke a key I think was leaked?

Go to /keys, click Revoke next to the affected key. Revocation is instant — subsequent requests carrying that key return 401 authentication_error at the edge, before any upstream call. Rotate clients to a new key, then delete the old row.

Is there an SLA?

Not in MVP. The service is best-effort; for real-time status you can subscribe to support@sandboxclaw.com status mails. Enterprise SLAs (uptime targets, dedicated capacity) are available on request after MVP.

Can I use token-hub from any country?

Yes, for every model in the catalog. Some upstream providers geo-restrict individual models — in those rare cases we surface 403 permission_denied or 502 upstream_error with the reason, and support can suggest an equivalent model in the same family.