Guides · 知识库

提示缓存是一种将对话中重复出现的系统提示、历史消息等前缀内容暂存于模型服务端，避免每次请求都重新计算的技术。本文解释其工作原理，以及在长多轮对话中如何通过缓存命中降低计费，并说明使用聚合站时的实际影响。

Prompt caching lets you reuse the unchanged prefix of a conversation across API calls, reducing token costs and latency. This article explains how caching works, why it matters for long multi-turn chats, and how to structure prompts to benefit on Claude and other models.

介绍聚合 API 中 model=auto 智能路由的机制、取舍与适用场景，帮助你在便利性、成本与可控性之间做出选择，并说明何时应固定具体模型。

Setting model=auto lets the platform choose a suitable model for each request based on availability and task characteristics. This guide explains how auto-routing works conceptually, the tradeoffs between routing and pinning, and when to explicitly specify a model.

本文以谨慎、事实性的视角，解释在使用 LLM API 聚合服务时，关于数据隐私与日志留存你需要了解的关键点。内容涵盖平台可能接触到的数据范围、透传给第三方模型供应商的信息，以及如何根据自身需求评估风险并采取缓解措施。不承诺绝对隐私，而是帮助你做出知情决策。

Learn what data the platform sees when you call models, how logging and retention work, and the privacy considerations of sending data to third-party models.

本文汇总了在使用 LLM API 聚合服务时可能遇到的常见错误，包括鉴权失败、请求限流、模型不存在等，并提供了排查思路与重试/退避的最佳实践。内容通用，不涉及具体错误码，旨在帮助开发者快速定位问题并提升应用稳定性。

A practical guide to handling common API errors like authentication failures, 429 rate limits, and model not found, with retry and backoff best practices.

介绍如何通过 OpenAI 兼容接口的 image_url 内容块向多模态模型发送图像输入，涵盖消息结构、图像传入方式（URL 与 base64）、多图与图文混排、常见注意事项，以及在不同模型间切换时的兼容性要点。

Learn how to send image inputs to multimodal models via the OpenAI-compatible format using image_url content blocks, including base64 encoding and URL references.

函数调用（Function Calling / Tool Use）让模型在对话中输出结构化的调用请求，由你的应用执行后再把结果回传。本文解释它是什么、什么时候值得用、请求里 tools 字段的基本结构，以及用聚合 API 时需要注意的兼容性差异。

Function calling lets models request structured calls to your own functions. This guide explains what it is, when to use it, and how to pass tools in a request, with a minimal example. It covers the request-response flow, best practices, and common pitfalls, so you can integrate external capabilities reliably with any supported model.

流式输出（SSE）让 LLM 响应边生成边返回，显著改善首字延迟与交互体验。本文说明如何在请求中开启 stream=true、SSE 分片长什么样、如何正确解析并拼接增量内容，并给出一个最小、框架无关的示例与常见坑。

Learn how to enable streaming responses from LLM APIs using Server-Sent Events (SSE). This guide explains the stream=true parameter, the structure of SSE chunks, and provides a minimal framework-agnostic code example for consuming streams in real time.

面向开发者的 Qwen 实操入门指南，介绍如何通过聚合站以 OpenAI 兼容格式最小化接入 Qwen 模型，包含环境准备、API 调用示例和计费说明。

A practical introduction to using Qwen models through the aggregator's OpenAI-compatible API, covering setup, a minimal code example, key model features, and use cases.

本文介绍如何通过 LLM API 聚合站快速上手 DeepSeek 模型，涵盖其擅长场景、获取 API key 的步骤，并给出一个兼容 OpenAI 格式的最小代码示例，帮助你用同一套代码调用多个模型。

A practical introduction to using DeepSeek models through an LLM API aggregator. Learn what DeepSeek excels at, how to access it via an OpenAI-compatible endpoint, and the basics of billing and credits.

按任务类型（写代码、写作、翻译、推理）给出选择 LLM 的概念性指南，概括 Claude、GPT、DeepSeek、Qwen 等模型家族的常见特点，并说明如何通过聚合 API 用同一套接口对比和切换模型。

A practical, conceptual guide to picking the right AI model for coding, writing, translation, and reasoning tasks. Learn how to match model strengths to your use case, using a single API key across many leading models.

介绍 LLM API 聚合站中保险池的概念与运作方式。解释为何鼓励用户一次性小额押金、共享 key 被封时如何获得赔付，以及健康检查在其中的作用。内容为概括性说明，不涉及具体数字与承诺。

A shared API key can be banned, disrupting access. An insurance pool, funded by a small one-time stake, provides a payout if that happens. Health checks monitor key status to catch bans early. This article explains the mechanics conceptually.

本文用通俗语言解释 LLM API 聚合站的双边定价机制：调用方按官方价 ×1.3 扣费，Key 贡献方按官方价 ×1.1（优质 ×1.2）获得 USDC 返利。帮助读者理解同一笔调用在自用视角下的资金流向与成本构成。

This article explains the transparent pricing model of an LLM API aggregator: users pay the official model price multiplied by 1.3, while key contributors receive credits at official price multiplied by 1.1 (or 1.2 for premium keys). It clarifies the two sides of a self-use call and how contributors can offset costs.

本文介绍在 LLM API 聚合站中，如何将贡献 key 获得的收益和账户余额提现成 Base 网络上的 USDC。内容涵盖提现资格、审批流程、少量提现费的概念说明，以及到账后的链上确认步骤。不涉及具体金额、价格或费率数字，仅提供流程性指导。

Learn how contributor earnings on our platform convert to USDC withdrawals on Base, including the approval process and a small withdrawal fee. Stay conceptual, no exact numbers.

本文介绍如何在 LLM API 聚合站使用 Base 网络上的 USDC 充值，包括生成充值地址、转账、自动入账及手动提交 txHash 兜底流程。

介绍如何通过 OpenAI 兼容的 /chat/completions 接口调用聚合站，在不改动现有 SDK 代码的前提下，仅修改 base_url 和 API key，即可切换至多个模型（如 Claude、GPT、DeepSeek 等）。内容涵盖请求格式、模型指定、流式响应以及常见注意事项。

Learn how to use the OpenAI-compatible /chat/completions endpoint with our API aggregator. Change one base URL and reuse your existing OpenAI or LangChain code to access multiple models with a single key, paying USDC on Base with no KYC.

本文分步介绍如何将 Claude Code 配置为通过聚合站使用 USDC 充值，从而调用 Claude 及其他模型。内容涵盖获取 API key、设置 ANTHROPIC_BASE_URL 和 ANTHROPIC_API_KEY 环境变量、充值方式、计费规则以及常见问题。

Learn how to configure Claude Code to use a crypto-paid API aggregator that supports the Anthropic API format. This guide covers setting the ANTHROPIC_BASE_URL environment variable, obtaining a USDC-funded API key, and verifying the setup. You'll also understand the pricing model: users pay official price ×1.3, and contributors earn credits at official price ×1.1 (premium ×1.2) in USDC.

一个 API key 即可调用多个主流 LLM 的实践指南，涵盖 OpenAI 兼容接口的使用方法，以及如何通过修改 model 字段在 Claude、GPT、DeepSeek、Qwen、GLM、Kimi 等模型间自由切换。

Learn how a single API key can give you access to multiple AI models like Claude, GPT, DeepSeek, Qwen, GLM, and Kimi. This article explains the OpenAI-compatible endpoint, how to switch models by changing the model field, and the benefits of using an aggregator with USDC on Base and no KYC.

本文解释为什么多数 LLM API 需要 KYC 和信用卡，以及无 KYC 聚合站如何用 USDC 充值绕过这些摩擦。内容覆盖从充值、余额到账到发起 API 调用的完整实操流程，并说明聚合站在多模型调用和成本结构上的特点。

Learn how to top up with USDC on Base: copy your deposit address, send USDC, and use the manual transaction hash fallback for exchange withdrawals if credits do not appear automatically.

Most official LLM APIs require KYC and a credit card, which adds friction for crypto users. A USDC-top-up aggregator on Base removes that friction by letting you deposit USDC, receive API credits, and call Claude, GPT, and other models with one API key.

让闲置的 API key 干活。分享到 llm.saylulu.com，别人每次调用你都能赚 USDC，key 被封还有保险赔付。

Put an idle API key to work. Share it on llm.saylulu.com and earn USDC whenever others use it — with insurance if it gets banned.