OpenAI-compatible API
An OpenAI API compatibility layer provided by Amazon Bedrock that allows existing OpenAI SDK invocation logic to migrate smoothly to Bedrock.
Core Facts
Timeline (last 90 days)
Xisu AI's OpenAI-compatible API exposes chat completions, embeddings, model list, and image generation endpoints; image generation is supported, but image editing and variations are not, and voice-related endpoints are not yet available.
An LLM API gateway (also called an AI gateway or proxy layer) is an intermediate service located between applications and large-model providers that exposes a standardized interface externally (commonly the OpenAI /chat/completions specification) and internally is responsible for translating and forwarding requests to different vendors' endpoints while uniformly handling authentication, billing, rate limiting, and logging.
A roughly 40-line Python wrapper can implement three features at once for LLM API calls: a hard timeout, exponential backoff retries, and model fallback.
Although different vendors' API interfaces are compatible, they still differ in model names, context windows, tool-calling formats, rate limits and billing methods, and the evaluation set should be re-run on the target vendor before the fallback solution goes live.
The developer chose Kimi based on three key requirements: reasoning based on numbers rather than feelings, reliable JSON output, and an OpenAI-compatible API. Integration took only one afternoon, using plain fetch, with no SDK lock-in, and unit tests were done with a mocked fetch (resolver/test/kimi.test.ts).
The sharing tool starts a gateway on the local machine, exposing the local model externally as an OpenAI-compatible API interface for other devices on the LAN to call.
Bedrock 提供两种调用 OpenAI 模型的接口：Converse API（原生接口，支持多轮对话、工具调用和流式响应）和 OpenAI 兼容 API（减少从 OpenAI 平台迁移的代码改动）。
If an interface exposes multiple models under the same OpenAI-compatible base URL, building a fallback chain is a low-cost insurance mechanism.
Gateways often name models differently from the original vendors, and subtle discrepancies (such as a period written as a hyphen) can cause 404 model not found errors.
All Facts (10)
The core design of the OpenAI API uses /v1/chat/completions as the main endpoint.
80%UnverifiedXisu AI's OpenAI-compatible API exposes chat completions, embeddings, model list, and image generation endpoints; image generation is supported, but image editing and variations are not, and voice-related endpoints are not yet available.
50%UnverifiedAn LLM API gateway (also called an AI gateway or proxy layer) is an intermediate service located between applications and large-model providers that exposes a standardized interface externally (commonly the OpenAI /chat/completions specification) and internally is responsible for translating and forwarding requests to different vendors' endpoints while uniformly handling authentication, billing, rate limiting, and logging.
50%UnverifiedA roughly 40-line Python wrapper can implement three features at once for LLM API calls: a hard timeout, exponential backoff retries, and model fallback.
50%UnverifiedAlthough different vendors' API interfaces are compatible, they still differ in model names, context windows, tool-calling formats, rate limits and billing methods, and the evaluation set should be re-run on the target vendor before the fallback solution goes live.
50%UnverifiedThe developer chose Kimi based on three key requirements: reasoning based on numbers rather than feelings, reliable JSON output, and an OpenAI-compatible API. Integration took only one afternoon, using plain fetch, with no SDK lock-in, and unit tests were done with a mocked fetch (resolver/test/kimi.test.ts).
50%UnverifiedThe sharing tool starts a gateway on the local machine, exposing the local model externally as an OpenAI-compatible API interface for other devices on the LAN to call.
50%UnverifiedBedrock 提供两种调用 OpenAI 模型的接口：Converse API（原生接口，支持多轮对话、工具调用和流式响应）和 OpenAI 兼容 API（减少从 OpenAI 平台迁移的代码改动）。
50%UnverifiedIf an interface exposes multiple models under the same OpenAI-compatible base URL, building a fallback chain is a low-cost insurance mechanism.
50%UnverifiedGateways often name models differently from the original vendors, and subtle discrepancies (such as a period written as a hyphen) can cause 404 model not found errors.
50%