Data flow and compliance

Know where prompts go before you send them.

Lowertoken is built for overseas developers who need cost-effective model routes and a clear answer to the first procurement question: what happens to my prompt data?

Request Path

Each chat completion request moves through a small number of systems. Lowertoken handles authentication, billing checks and routing. The selected upstream model provider processes the prompt and completion under its own terms.

1. ClientYour app sends an OpenAI-compatible request to api.lowertoken.com.
2. GatewayLowertoken validates the key, balance and rate limits.
3. Route policyThe gateway selects a provider route based on model name and availability.
4. UpstreamThe provider endpoint processes prompt content and returns a response.
5. ResponseThe response streams back through Lowertoken to your client.

Upstream Selection

Phase 1 prioritizes lower-cost domestic China-route capacity and labels it clearly. Provider regions, retention and training policies may vary by upstream account, endpoint and model family.

Provider route Model family Preferred access posture Launch status
Tencent TokenHubDeepSeek, Kimi, MiniMax, Qwen, GLM, Hy3Primary prepaid China-route aggregator. Requests are sent to Tencent TokenHub and its selected upstream model route.Primary live
DeepSeek official APIDeepSeek chat and reasonerOfficial China-route endpoint kept as a direct specialist or fallback route.Live
Alibaba Cloud Model Studio / DashScopeQwen qwen3.6-plusCurrent live route uses the DashScope endpoint. Treat this as a China-route provider until an Alibaba Cloud International account and region are enabled.Live
SiliconFlow ChinaAggregated efficient modelsChina-route aggregator used for long-tail coverage and fallback routing.Live
Moonshot / Kimi ChinaKimiChina-route endpoint kept for direct Kimi models and route comparison.Live
Zhipu / GLM ChinaGLMChina-route endpoint kept for direct GLM models and fallback routing.Live

Retention and Training Use

Prompt bodies

Lowertoken does not intentionally persist prompt or completion bodies in application logs beyond transient buffers needed to stream the response. Prompt content is transmitted to the selected upstream provider to generate the response.

Request metadata

We retain operational metadata such as account, key, model, token counts, timestamps, status codes and billing events for abuse prevention, support and accounting.

Model training

Lowertoken does not use customer prompts to train models. Operational logs are designed to avoid prompt and completion text. Upstream providers may apply their own processing and retention practices.

Customer Controls

Control Phase 1 behavior Team option
Provider allowlistPublic routes listed in console.Manual allowlist review for teams.
Provider denylistModel-level route selection.Manual denylist review for sensitive workloads.
Usage limitsPer-key limits and prepaid balance.Higher limits after manual review.
Data questionsSupport through console contact.Routing summary before production use.

Data Requests

Customers can request account deletion, metadata export or routing clarification through the support contact published in the console. For regulated workloads, review upstream provider terms before sending production data.