Service levels & support服务等级与支持
Last updated: November 2026最后更新：2026 年 11 月
Uptime
Target: 99.5% monthly availability of the gateway, meaning the API endpoints accept and forward requests. It is measured by our external monitors and published on the status page, together with the upstream provider's availability, which is shown separately. Failed requests are never billed, whatever the cause.
Service credits (wholesale agreements)
| Gateway availability in a month | Credit |
|---|---|
| Below 99.5% | 5% of that month's consumption |
| Below 99.0% | 10% (the maximum) |
Credits never exceed 10% of that month's consumption. They are added to your balance; claim within 30 days of the month's end. Minutes when the upstream provider itself is unavailable are shown on the status page but are outside our control and not credited; minutes of announced maintenance are excluded.
Maintenance
Announced at least 48 hours ahead on the status page, scheduled between 02:00 and 06:00 UTC, and kept under two hours a month in total.
Support
- Channel: email, using the contact shown on this site. Quote the
x-relay-request-idheader from the failing response. - First response: within one business day for self-serve accounts; within 24 hours for wholesale accounts, in English or Chinese (Toronto time, Monday to Friday).
- Incidents: acknowledged within one hour between 09:00 and 23:00 Toronto time; status page updated at least every 30 minutes until resolved; a short post-incident note within three business days for wholesale accounts.
Model changes
We mirror the upstream provider's lifecycle. When a model id is scheduled for retirement we give at least 30 days' notice by email and on the status page, with the replacement id. New models are added to the price page as they become available upstream.
Rate limits
Each key has a requests-per-minute limit shown in the console; wholesale limits are agreed in the plan. Over the limit, the API returns 429 with a retry-after header. Tell us three business days ahead of a launch, campaign or expected peak and we will raise the limit.
Data
Prompts, images and videos are passed to the upstream provider and are not written to our database; see the Privacy Policy. Usage records (time, model, units, cost, status, latency) are kept for billing and are available in the console and on the monthly statement.
可用性
目标：网关月度可用性 99.5%，即 API 端点能接收并转发请求。由我们的外部监控测量，并在状态页公布；上游厂商自身的可用性单独显示。无论原因，失败的请求一律不收费。
服务补偿（批发协议）
| 当月网关可用性 | 补偿 |
|---|---|
| 低于 99.5% | 当月消耗的 5% |
| 低于 99.0% | 10%（上限） |
补偿总额不超过当月消耗的 10%。
补偿以余额形式发放，需在当月结束后 30 天内提出。上游厂商自身不可用的时间会显示在状态页上，但不在我们控制范围内，不予补偿；已公告的维护时间不计入。
维护
至少提前 48 小时在状态页公告，安排在 UTC 02:00～06:00 之间，每月合计不超过两小时。
支持
- 渠道：邮件，使用本站显示的联系方式。请附上出错响应里的
x-relay-request-id。 - 首次响应：自助账号一个工作日内；批发账号 24 小时内，中英文均可（多伦多时间，周一至周五）。
- 故障：多伦多时间 09:00～23:00 之间一小时内确认；状态页至少每 30 分钟更新一次直到解决；批发账号在三个工作日内收到简短的事后说明。
模型变更
我们跟随上游厂商的生命周期。某个模型 ID 计划下线时，我们至少提前 30 天通过邮件和状态页通知，并给出替代 ID。新模型在上游可用后加入价格页。
限流
每个密钥都有每分钟请求数上限，显示在控制台；批发限额在套餐里约定。超限时 API 返回 429 并附 retry-after。上线、活动或预计有峰值前提前 3 个工作日告诉我们，我们会提高限额。
数据
提示词、图片和视频只转给上游厂商，不写入我们的数据库，见隐私政策。用量记录（时间、模型、用量、费用、状态、耗时）用于计费，可在控制台和月度对账单中查看。