About us

1. What we build

InOneAPI is a production-oriented AI model API aggregation and smart routing platform. We put text, image, video, and audio models from China's top model providers behind one API that is compatible with the OpenAI and Anthropic SDKs, so your application selects a provider by business priority instead of maintaining an integration, a key, and a reconciliation process for each one.

We connect only to China's top-tier model providers. We do not add low-cost channels whose origin and quality we cannot verify.

InOneAPI does not develop large models and is not the party that serves model inference. Requests you send are completed by the provider channel you choose or the one routing selects. A provider's retention policy, content review rules, and regional compliance follow its own terms.

2. What we focus on

Reliability

Where possible, the same model is available through more than one provider. The gateway tracks changes in latency, throughput, and error rate, adjusts paths between the available channels for the same model, and switches on failure, which reduces the impact of depending on a single provider. We do not make an unconditional availability promise: routing and failover reduce interruption risk and do not replace retry and degradation handling in your application.

Cost

Price-first routing selects the lowest-priced available provider for the same model. Speed-first and balanced modes are available when your priorities differ. Each request is billed at the price of the provider that actually served it, at provider prices with no markup. Model, token usage, and cost per request are reviewable in the console.

Privacy

Zero content logging is the platform's standard operating mode, not an optional setting. Prompts and model responses are forwarded, streamed, and counted in memory only, and are never written to a database. Request metadata used for billing and diagnostics is kept separate from content and never includes prompt or completion text.

3. Platform capabilities

  • Smart routing: price-first, speed-first, and balanced modes that select paths using price, time to first token, throughput, error rate, and availability.
  • Gateway performance: a Rust gateway keeps persistent HTTP/2 Keep-Alive connection pools to upstream providers to improve time to first token.
  • Built-in LLMOps trace: request-level X-Gateway-Trace-ID with timing, retry, token usage, and input/output diagnostics.
  • Abnormal traffic control: detection of runaway agents, retry storms, and leaked keys, with active circuit breaking and WeChat or email alerts to administrators.
  • Transparent billing: local token calculation aligned to official provider algorithms, billed per request and verifiable line by line.
  • Team governance: workspace, project, member, and sub-key access boundaries with budget limits and model allowlists.
  • Playground: validate model behavior and parameters in the Playground and model experience pages before you write production code.

4. Accounts and billing

Every account is assigned one immutable currency and a matching wallet at registration — USD, recharged through Stripe — and changing the interface language never changes it. Prices and invoices are not converted after the fact: each model carries an independent USD and CNY price.

5. Who we serve

  • AI application and agent developers who need China's models through one integration rather than several.
  • Engineering and platform teams that need to distribute keys and control budgets and model scope across members and projects.
  • Businesses that need verifiable cost records and no content retention.

6. Where to go next

About us · InOneAPI