[{"data":1,"prerenderedAt":1375},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"compare-openrouter-vs-litellm":8,"compare-a-openrouter":9,"compare-b-litellm":617},{"tools":4,"reviews":5},98,27,{"tools":4,"reviews":5,"playbooks":7,"news":5},22,null,{"id":10,"title":11,"alternatives":12,"api_compatible":8,"body":17,"category":538,"chinese_friendly":528,"cover":539,"description":540,"domestic":541,"extension":542,"faq":543,"free":541,"github":8,"languages":556,"lastVerified":8,"meta":559,"models":8,"navigation":560,"notSuitable":8,"opensource":541,"path":561,"pillar":562,"platforms":563,"priceTable":565,"pricing":582,"published":583,"relatedPlaybooks":584,"relatedReviews":587,"score":589,"self_host":541,"seo":592,"seoTitle":593,"slug":594,"sources":595,"stem":605,"suitable":8,"tagline":606,"tags":607,"updated":598,"verdict":614,"website":615,"__hash__":616},"tools\u002Ftools\u002Fcoding\u002Fapi\u002Fopenrouter.md","OpenRouter",[13,14,15,16],"coding\u002Fapi\u002Flitellm","coding\u002Fapi\u002Fone-api","coding\u002Fapi\u002Fportkey","coding\u002Fapi\u002Fhelicone",{"type":18,"value":19,"toc":523},"minimark",[20,25,29,32,35,116,119,150,156,160,165,182,187,211,214,247,250,385,388,426,430,459,462,489,492],[21,22,24],"h2",{"id":23},"tldr","TL;DR",[26,27,28],"p",{},"OpenRouter 是 AI API 聚合先驱——一个 OpenAI 兼容 endpoint + 一个 Key，调 60+ 厂商 300+ 模型，5M+ 用户。模型 passthrough 定价 0 markup，但充值有 5.5% 信用卡平台费（$0.80 最低）+ BYOK 月 100 万请求后 5%。自动 fallback \u002F playground \u002F 海量模型选型是杀手锏。国内直连延迟 1500–3000ms + 不支持支付宝是硬伤。",[26,30,31],{},"适合：海外开发者；多模型选型 \u002F 横评；生产 fallback 高可用；零迁移成本（OpenAI SDK 改 baseURL）。不适合：国内对延迟敏感的 AI 编程工具（用 OfoxAI \u002F 自建中转）；要支付宝 \u002F 微信 \u002F 银联；要 self-host（用 one-api \u002F LiteLLM）。",[21,33,34],{"id":34},"核心能力",[36,37,38,46,57,70,80,86,92,98,104,110],"ul",{},[39,40,41,45],"li",{},[42,43,44],"strong",{},"60+ 厂商 300+ 模型","：OpenAI \u002F Anthropic \u002F Google \u002F Meta \u002F DeepSeek \u002F MiniMax \u002F Mistral \u002F Cohere \u002F xAI \u002F NVIDIA \u002F Qwen \u002F 字节豆包 等",[39,47,48,51,52,56],{},[42,49,50],{},"OpenAI 兼容 endpoint","：",[53,54,55],"code",{},"https:\u002F\u002Fopenrouter.ai\u002Fapi\u002Fv1","，SDK 改 baseURL 即用",[39,58,59,51,62,65,66,69],{},[42,60,61],{},"模型 ID 格式",[53,63,64],{},"provider\u002Fmodel-name","（如 ",[53,67,68],{},"anthropic\u002Fclaude-sonnet-4.6","）",[39,71,72,75,76,79],{},[42,73,74],{},"自动 fallback","：请求里列 ",[53,77,78],{},"models: [...]","，前者失败自动切后者",[39,81,82,85],{},[42,83,84],{},"Playground","：5 模型并排测同 prompt + 成本 + 延迟 + 输出",[39,87,88,91],{},[42,89,90],{},"BYOK","：自带 OpenAI \u002F Anthropic Key，OpenRouter 只做网关；前 100 万 req\u002F月免费，超 5% fee",[39,93,94,97],{},[42,95,96],{},"Passthrough 定价","：模型按厂商官方价",[39,99,100,103],{},[42,101,102],{},"免费模型","：Llama 4 Scout \u002F DeepSeek V4 Flash \u002F Qwen3 Coder \u002F Gemma 3 等，每日 5\u002F200 次",[39,105,106,109],{},[42,107,108],{},"Provider ranking \u002F SLA","：dashboard 看每家 provider 延迟 + 错误率",[39,111,112,115],{},[42,113,114],{},"Rankings 页","：模型按 token 消耗排名（市场热度参考）",[21,117,118],{"id":118},"价格",[36,120,121,127,133,139,145],{},[39,122,123,126],{},[42,124,125],{},"Free","：$1 credit + 免费模型 5\u002F日 + 标准限流",[39,128,129,132],{},[42,130,131],{},"Pay-as-you-go","：模型 passthrough + 5.5% 卡费（最低 $0.80）+ 充值 $5 后免费模型涨到 200\u002F日",[39,134,135,138],{},[42,136,137],{},"Team","：$99\u002F月 + Priority routing + 99.9% SLA",[39,140,141,144],{},[42,142,143],{},"Enterprise","：Custom 专属基础设施",[39,146,147,149],{},[42,148,90],{},"：前 100 万 req\u002F月免费，超出 5%",[151,152,153],"blockquote",{},[26,154,155],{},"实际总成本 ≈ 模型价 × (1 + 5.5%) + BYOK 5% × 月 >1M 部分。预算按 5–7% overhead 算。",[21,157,159],{"id":158},"实测小型-saas-后端-海外节点","实测（小型 SaaS 后端 \u002F 海外节点）",[26,161,162],{},[42,163,164],{},"亮点：",[36,166,167,170,173,176,179],{},[39,168,169],{},"5 分钟接入，OpenAI SDK 改两行配置",[39,171,172],{},"Playground 横评新模型省一大堆 boilerplate",[39,174,175],{},"自动 fallback 救命：Anthropic 4 小时宕机时应用零中断",[39,177,178],{},"模型 ID 提示器 + Rankings 看市场热度方便选型",[39,180,181],{},"海外节点延迟 300–600ms，CDN 体感好",[26,183,184],{},[42,185,186],{},"踩坑（国内）：",[36,188,189,192,195,198,205,208],{},[39,190,191],{},"国内直连 1500–3000ms + 高峰超时率明显，Cursor \u002F Claude Code 高频调用体感糟",[39,193,194],{},"不支持支付宝 \u002F 微信 \u002F 银联，只 Visa\u002FMastercard \u002F USDC",[39,196,197],{},"信用卡费 5.5% 最低 $0.80，\u003C$15 充值实际打到 10%+",[39,199,200,201,204],{},"模型 ID 点号分隔（",[53,202,203],{},"claude-sonnet-4.6","）易跟连字符混",[39,206,207],{},"Claude Code 默认走 Anthropic 原生协议，OpenRouter 适配偶有不一致",[39,209,210],{},"「免费模型」rate limit 后可能静默切付费",[21,212,213],{"id":213},"上手",[215,216,217,220,223,226,232,238,244],"ol",{},[39,218,219],{},"openrouter.ai → Google \u002F GitHub OAuth 注册",[39,221,222],{},"Keys → 创建 API Key（sk-or-xxx）",[39,224,225],{},"充值 $10+（Visa \u002F Mastercard \u002F USDC）→ 解锁更高免费配额",[39,227,228,229],{},"代码：",[53,230,231],{},"OpenAI(baseURL='https:\u002F\u002Fopenrouter.ai\u002Fapi\u002Fv1', api_key=KEY)",[39,233,234,235,237],{},"模型用 ",[53,236,64],{}," 格式",[39,239,240,241],{},"加 fallback：",[53,242,243],{},"extra_body={\"models\": [\"anthropic\u002Fclaude-sonnet-4.6\", \"openai\u002Fgpt-5.4\"]}",[39,245,246],{},"Playground 横评再批量切",[21,248,249],{"id":249},"对比",[251,252,253,274],"table",{},[254,255,256],"thead",{},[257,258,259,263,265,268,271],"tr",{},[260,261,262],"th",{},"维度",[260,264,11],{},[260,266,267],{},"LiteLLM",[260,269,270],{},"One-API",[260,272,273],{},"Portkey",[275,276,277,294,311,326,340,353,369],"tbody",{},[257,278,279,283,286,289,292],{},[280,281,282],"td",{},"形态",[280,284,285],{},"SaaS",[280,287,288],{},"OSS + SaaS",[280,290,291],{},"OSS",[280,293,285],{},[257,295,296,299,302,305,308],{},[280,297,298],{},"模型数",[280,300,301],{},"300+",[280,303,304],{},"100+",[280,306,307],{},"30+",[280,309,310],{},"250+",[257,312,313,316,319,322,324],{},[280,314,315],{},"自托管",[280,317,318],{},"❌",[280,320,321],{},"✅",[280,323,321],{},[280,325,143],{},[257,327,328,331,333,336,338],{},[280,329,330],{},"中文支付",[280,332,318],{},[280,334,335],{},"–",[280,337,321],{},[280,339,335],{},[257,341,342,345,347,349,351],{},[280,343,344],{},"Fallback",[280,346,321],{},[280,348,321],{},[280,350,321],{},[280,352,321],{},[257,354,355,358,361,364,366],{},[280,356,357],{},"国内延迟",[280,359,360],{},"1500–3000ms",[280,362,363],{},"自托管自定",[280,365,363],{},[280,367,368],{},"海外为主",[257,370,371,374,377,380,382],{},[280,372,373],{},"成本",[280,375,376],{},"模型 + 5.5%\u002FBYOK 5%",[280,378,379],{},"自付服务器",[280,381,379],{},[280,383,384],{},"按 logs 计费",[21,386,387],{"id":387},"避坑",[36,389,390,396,402,408,414,420],{},[39,391,392,395],{},[42,393,394],{},"国内别裸用","：1500ms+ 起、超时率高，主链路上 OfoxAI \u002F 自建 one-api 中转，OpenRouter 当冷门模型补",[39,397,398,401],{},[42,399,400],{},"预算多算 5–7%","：5.5% 信用卡 + BYOK 5% + 偶发汇率",[39,403,404,407],{},[42,405,406],{},"免费模型生产慎用","：rate limit 后可能静默 fallback 付费，账单失控",[39,409,410,413],{},[42,411,412],{},"模型 ID 格式坑","：点号 vs 连字符要测；新模型 ID 改名要监控",[39,415,416,419],{},[42,417,418],{},"fallback 别堆太多","：5+ fallback 一次失败延迟叠加可能 10s+，前 2 个即可",[39,421,422,425],{},[42,423,424],{},"Claude Code 原生模式建议直连","：OpenRouter 走 OpenAI 兼容协议有时与原生 Anthropic 协议有出入",[21,427,429],{"id":428},"适合-不适合","适合 \u002F 不适合",[36,431,432,435,438,441,444,447,450,453,456],{},[39,433,434],{},"✅ 海外开发者 \u002F 跨境业务",[39,436,437],{},"✅ 多模型选型 \u002F 横评",[39,439,440],{},"✅ 生产 fallback 高可用",[39,442,443],{},"✅ 零迁移成本（OpenAI SDK 改 baseURL）",[39,445,446],{},"✅ BYOK 模式给团队统一 Key 管理",[39,448,449],{},"❌ 国内 AI 编程工具（Cursor \u002F Claude Code）高频调用",[39,451,452],{},"❌ 要支付宝 \u002F 微信 \u002F 银联",[39,454,455],{},"❌ Self-host \u002F 完全控制数据",[39,457,458],{},"❌ 极致低延迟实时应用",[21,460,461],{"id":461},"相关阅读",[36,463,464,471,477,483],{},[39,465,466],{},[467,468,470],"a",{"href":469},"\u002Ftools\u002Fcoding\u002Fapi\u002Flitellm","LiteLLM 评测",[39,472,473],{},[467,474,476],{"href":475},"\u002Ftools\u002Fcoding\u002Fapi\u002Fone-api","One-API 评测",[39,478,479],{},[467,480,482],{"href":481},"\u002Ftools\u002Fcoding\u002Fapi\u002Fportkey","Portkey 评测",[39,484,485],{},[467,486,488],{"href":487},"\u002Ftools\u002Fcoding\u002Fapi\u002Fhelicone","Helicone 评测",[21,490,491],{"id":491},"来源",[215,493,494,502,509,516],{},[39,495,496,497],{},"OfoxAI — OpenRouter 完全指南 2026（国内延迟实测 + 替代方案）",[467,498,499],{"href":499,"rel":500},"https:\u002F\u002Fofox.ai\u002Fzh\u002Fblog\u002Fopenrouter-complete-guide-china-developers-2026",[501],"nofollow",[39,503,504,505],{},"OfoxAI — OpenRouter Pricing 2026 Hidden Markup Breakdown ",[467,506,507],{"href":507,"rel":508},"https:\u002F\u002Fofox.ai\u002Fblog\u002Fopenrouter-pricing-hidden-markup-breakdown-2026\u002F",[501],[39,510,511,512],{},"DigitalApplied — OpenRouter June 2026 New Models & Pricing Roundup ",[467,513,514],{"href":514,"rel":515},"https:\u002F\u002Fwww.digitalapplied.com\u002Fblog\u002Fopenrouter-new-models-june-2026-roundup-pricing-rankings",[501],[39,517,518,519],{},"AIToolBox — OpenRouter Review 2026 ",[467,520,521],{"href":521,"rel":522},"https:\u002F\u002Fwww.aitoolbox.hk\u002Ftools\u002Fopen-router\u002F",[501],{"title":524,"searchDepth":525,"depth":525,"links":526},"",3,[527,529,530,531,532,533,534,535,536,537],{"id":23,"depth":528,"text":24},2,{"id":34,"depth":528,"text":34},{"id":118,"depth":528,"text":118},{"id":158,"depth":528,"text":159},{"id":213,"depth":528,"text":213},{"id":249,"depth":528,"text":249},{"id":387,"depth":528,"text":387},{"id":428,"depth":528,"text":429},{"id":461,"depth":528,"text":461},{"id":491,"depth":528,"text":491},"api","\u002Fimg\u002Ftools\u002Fopenrouter.webp","OpenRouter 真实评测：AI API 聚合先驱，60+ 厂商 300+ 模型一个 OpenAI 兼容 endpoint。passthrough 定价 0 markup + 5.5% 信用卡平台费（$0.80 最低）+ BYOK 5% 月千万请求外。Pay-as-you-go \u002F Team $99\u002F月 SLA \u002F Enterprise。国内直连 1500–3000ms 高峰超时多。",false,"md",[544,547,550,553],{"q":545,"a":546},"国内能用吗？","技术上能（OpenAI SDK 改 baseURL），但服务器在海外、无国内节点：实测首 Token 延迟代理 600–1200ms \u002F 直连 1500–3000ms 且不稳，高峰超时率高。AI 编程类高频调用（Cursor \u002F Claude Code \u002F Cline）国内体验糟糕，主流方案是用 OfoxAI \u002F 自建 one-api\u002Fnew-api 中转。",{"q":548,"a":549},"真的零 markup 吗？","模型本身按厂商官方价 passthrough，技术上 0 markup。但充值有两层平台费：信用卡 5.5%（最低 $0.80，小额充值实际打到 10–20%）、BYOK 月 100 万请求后 5%。预算时按上层多算 5–7%。",{"q":551,"a":552},"自动 fallback 怎么用？","请求里加 `models: ['anthropic\u002Fclaude-sonnet-4.6', 'openai\u002Fgpt-5.4']`，前者失败 \u002F 超时 \u002F 限流时自动切到后者。生产 24h+ 不间断 AI 服务几乎刚需——Anthropic 2026 年 5 月 4 小时宕机，用 fallback 的应用零中断。",{"q":554,"a":555},"免费模型怎么用？","模型列表筛选 $0 价（如 Llama 4 Scout \u002F DeepSeek V4 Flash \u002F Qwen3 Coder \u002F Gemma 3）。未充值每日 5 次，充 $5+ 涨到 200 次\u002F日，附带 20 req\u002Fmin。生产别依赖免费模型，限流 \u002F 排队 \u002F 暗切付费会让计费失控。",[557,558],"en","zh",{},true,"\u002Ftools\u002Fcoding\u002Fapi\u002Fopenrouter","coding",[538,564],"sdk",[566,570,574,578],{"plan":125,"price":567,"features":568,"notes":569},"$0","$1 credit + 免费模型 5\u002F日 + 标准限流","不需要信用卡",{"plan":131,"price":571,"features":572,"notes":573},"Passthrough + 5.5% 卡费","300+ 模型 + 标准限流 + 充 $5 后免费模型 200\u002F日 + BYOK","信用卡平台费最低 $0.80",{"plan":137,"price":575,"features":576,"notes":577},"$99\u002F月","Priority routing + 99.9% SLA + 更高限流","面向生产应用",{"plan":143,"price":579,"features":580,"notes":581},"Custom","专属基础设施 + 自定义合同","大规模生产","免费 $1 credit + 5\u002F日 → 200\u002F日（充 $5+） \u002F Pay-as-you-go passthrough + 5.5% 卡费 \u002F Team $99\u002F月 SLA \u002F Enterprise","2026-06-19",[585,586],"onboarding\u002Frag-pipeline-build","onboarding\u002Fclaude-code-getting-started",[588],"llm-gateway-comparison",{"power":590,"ux":591,"price":591,"cn_support":528,"stability":591},5,4,{"title":11,"description":540},"OpenRouter 评测 2026：多模型 API 统一网关，对比各大 AI API","coding\u002Fapi\u002Fopenrouter",[596,599,601,603],{"name":597,"url":499,"accessed":598},"OfoxAI — OpenRouter 完全指南 2026","2026-06-24",{"name":600,"url":507,"accessed":598},"OfoxAI — OpenRouter Pricing Hidden Markup",{"name":602,"url":514,"accessed":598},"DigitalApplied — OpenRouter June 2026 Roundup",{"name":604,"url":521,"accessed":598},"AIToolBox — OpenRouter Review 2026","tools\u002Fcoding\u002Fapi\u002Fopenrouter","AI API 聚合先驱——一个 Key + 一个 endpoint 调 300+ 模型，自动 fallback，OpenAI 兼容",[608,609,610,611,612,613],"api-gateway","aggregator","openai-compatible","fallback","multi-model","byok","AI API 聚合的全球事实标准——300+ 模型 \u002F 一个 Key \u002F 自动 fallback。海外开发者首选；国内开发者延迟 1500–3000ms + 不支持支付宝是硬伤，主走 OfoxAI \u002F one-api 自建中转。","https:\u002F\u002Fopenrouter.ai","pc5ZNJB9xmKwAiAzBCr-JjF35aQPkc37A3rZxBWuxtQ",{"id":618,"title":267,"alternatives":619,"api_compatible":8,"body":620,"category":538,"chinese_friendly":525,"cover":1321,"description":1322,"domestic":541,"extension":542,"faq":1323,"free":541,"github":8,"languages":1336,"lastVerified":8,"meta":1337,"models":8,"navigation":560,"notSuitable":8,"opensource":560,"path":469,"pillar":562,"platforms":1338,"priceTable":1341,"pricing":1350,"published":583,"relatedPlaybooks":1351,"relatedReviews":1352,"score":1353,"self_host":560,"seo":1354,"seoTitle":1355,"slug":13,"sources":1356,"stem":1365,"suitable":8,"tagline":1366,"tags":1367,"updated":598,"verdict":1372,"website":1373,"__hash__":1374},"tools\u002Ftools\u002Fcoding\u002Fapi\u002Flitellm.md",[594,14,15,16],{"type":18,"value":621,"toc":1309},[622,624,631,634,636,731,733,748,753,757,761,784,789,813,817,1016,1018,1158,1160,1224,1226,1252,1254,1273,1275,1305],[21,623,24],{"id":23},[26,625,626,627,630],{},"LiteLLM 是开源 LLM 网关里事实标准：MIT 协议，BerriAI 维护，双形态——SDK（",[53,628,629],{},"pip install","，单进程嵌入）+ Proxy（Docker Compose + Postgres，团队级网关）。100+ 厂商统一 OpenAI 接口，虚拟 Key + 团队预算 + 三类 fallback（错误\u002F政策\u002F上下文）+ 成本追踪 + 内置 admin UI。2026-03 供应链事件后 v1.83+ 强化镜像验证，生产固定签名版本。",[26,632,633],{},"适合：中大型团队 \u002F 合规 \u002F 想完全控数据 + BYOK；微服务架构需要语言无关的 OpenAI 网关；要把成本治理 + 观测做到自建。不适合：不想运维（SaaS 走 OpenRouter \u002F Portkey）；中文支付 \u002F 业务运营（走 one-api \u002F new-api）；纯个人项目（pip 装 SDK 已够，不需要 Proxy）。",[21,635,34],{"id":34},[36,637,638,644,653,662,671,680,686,692,701,707,713,719,725],{},[39,639,640,643],{},[42,641,642],{},"100+ providers","：OpenAI \u002F Anthropic \u002F Google \u002F AWS Bedrock \u002F Azure \u002F vLLM \u002F Ollama \u002F Together \u002F HuggingFace 等",[39,645,646,649,650],{},[42,647,648],{},"统一 OpenAI 接口","：所有模型走 ",[53,651,652],{},"\u002Fv1\u002Fchat\u002Fcompletions",[39,654,655,51,658,661],{},[42,656,657],{},"SDK 模式",[53,659,660],{},"from litellm import completion","，适合嵌入",[39,663,664,667,668,69],{},[42,665,666],{},"Proxy 模式","：HTTP 服务 + Postgres + admin UI（",[53,669,670],{},"\u002Fui",[39,672,673,51,676,679],{},[42,674,675],{},"虚拟 Key",[53,677,678],{},"\u002Fkey\u002Fgenerate"," 给团队 \u002F 服务签发独立 Key + 预算 + 模型白名单",[39,681,682,685],{},[42,683,684],{},"三类 fallback","：错误 \u002F 内容政策 \u002F context window",[39,687,688,691],{},[42,689,690],{},"重试 + 超时 + cooldown","：完整 SRE 配套",[39,693,694,51,697,700],{},[42,695,696],{},"Cost tracking",[53,698,699],{},"\u002Fglobal\u002Fspend\u002Freport"," + admin UI 看每团队 \u002F Key \u002F 模型成本",[39,702,703,706],{},[42,704,705],{},"回调","：Langfuse \u002F Prometheus \u002F Slack 一键挂载",[39,708,709,712],{},[42,710,711],{},"Guardrails","：Presidio PII masking \u002F 自定义内容检查",[39,714,715,718],{},[42,716,717],{},"缓存","：Redis 内置 + 语义缓存",[39,720,721,724],{},[42,722,723],{},"Routing strategy","：cost-based \u002F latency-based \u002F round-robin",[39,726,727,730],{},[42,728,729],{},"MIT 协议","：完全自由商用 + 修改",[21,732,118],{"id":118},[36,734,735,740,745],{},[39,736,737,739],{},[42,738,291],{},"：$0；自托管成本 = 1 台 Postgres + 1 台 LiteLLM container（~2GB RAM）",[39,741,742,744],{},[42,743,143],{},"：Custom；SSO \u002F SAML \u002F 审计 \u002F SLA \u002F on-prem 部署支持",[39,746,747],{},"模型成本走各厂商直接结算（BYOK）",[151,749,750],{},[26,751,752],{},"小规模总成本：1 台 2 核 4G VPS 跑 Postgres + LiteLLM 月 $20–30，团队 10 人完全够。",[21,754,756],{"id":755},"实测10-人-saas-微服务架构","实测（10 人 SaaS \u002F 微服务架构）",[26,758,759],{},[42,760,164],{},[36,762,763,766,769,772,775,778,781],{},[39,764,765],{},"Docker Compose 40 分钟拉起完整生产栈",[39,767,768],{},"虚拟 Key + 预算让微服务团队成本归因清晰",[39,770,771],{},"三类 fallback 配齐后可用率从 99.2% → 99.8%",[39,773,774],{},"admin UI 看每团队每天成本省了一堆自研 dashboard",[39,776,777],{},"Postgres + master key 模式，密钥 + 配置一致性高",[39,779,780],{},"与 Langfuse 集成做 trace + cost 双视角",[39,782,783],{},"多语言客户端（Python \u002F Node \u002F Go \u002F Rust）走同一 endpoint 体验一致",[26,785,786],{},[42,787,788],{},"踩坑：",[36,790,791,798,801,804,810],{},[39,792,793,794,797],{},"2026-03 供应链事件让团队对 ",[53,795,796],{},":latest"," tag 警惕，生产必须固定签名版本（如 v1.85.0）",[39,799,800],{},"Postgres salt key 一旦生成不能轮换，初始化前要谨慎备份",[39,802,803],{},"admin UI 早期版本英文为主，中文文档少",[39,805,806,807,809],{},"模型 ID 命名复杂（",[53,808,64],{},"），各厂商命名规则不一",[39,811,812],{},"自托管 = 自付运维（Postgres backup \u002F 升级 \u002F 监控）",[21,814,816],{"id":815},"上手proxy-模式","上手（Proxy 模式）",[818,819,823],"pre",{"className":820,"code":821,"language":822,"meta":524,"style":524},"language-bash shiki shiki-themes github-light github-dark","mkdir llm-gateway && cd llm-gateway\nopenssl rand -hex 32 > .master_key\nopenssl rand -hex 32 > .salt_key\n\n# docker-compose.yml 拉 ghcr.io\u002Fberriai\u002Flitellm:v1.85.0\n# config.yaml 写 model_list \u002F fallbacks \u002F litellm_settings\n\ndocker compose up -d\n\n# 生成虚拟 Key\ncurl -X POST http:\u002F\u002Flocalhost:4000\u002Fkey\u002Fgenerate \\\n  -H \"Authorization: Bearer $LITELLM_MASTER_KEY\" \\\n  -d '{\"models\":[\"gpt-5.4\",\"claude-sonnet-4.6\"],\"max_budget\":100}'\n\n# 业务方调用\ncurl http:\u002F\u002Flocalhost:4000\u002Fv1\u002Fchat\u002Fcompletions \\\n  -H \"Authorization: Bearer sk-xxx\" \\\n  -d '{\"model\":\"gpt-5.4\",\"messages\":[...]}'\n","bash",[53,824,825,849,870,885,890,896,902,907,922,927,933,951,968,977,982,988,998,1008],{"__ignoreMap":524},[826,827,830,834,838,842,846],"span",{"class":828,"line":829},"line",1,[826,831,833],{"class":832},"sScJk","mkdir",[826,835,837],{"class":836},"sZZnC"," llm-gateway",[826,839,841],{"class":840},"sVt8B"," && ",[826,843,845],{"class":844},"sj4cs","cd",[826,847,848],{"class":836}," llm-gateway\n",[826,850,851,854,857,860,863,867],{"class":828,"line":528},[826,852,853],{"class":832},"openssl",[826,855,856],{"class":836}," rand",[826,858,859],{"class":844}," -hex",[826,861,862],{"class":844}," 32",[826,864,866],{"class":865},"szBVR"," >",[826,868,869],{"class":836}," .master_key\n",[826,871,872,874,876,878,880,882],{"class":828,"line":525},[826,873,853],{"class":832},[826,875,856],{"class":836},[826,877,859],{"class":844},[826,879,862],{"class":844},[826,881,866],{"class":865},[826,883,884],{"class":836}," .salt_key\n",[826,886,887],{"class":828,"line":591},[826,888,889],{"emptyLinePlaceholder":560},"\n",[826,891,892],{"class":828,"line":590},[826,893,895],{"class":894},"sJ8bj","# docker-compose.yml 拉 ghcr.io\u002Fberriai\u002Flitellm:v1.85.0\n",[826,897,899],{"class":828,"line":898},6,[826,900,901],{"class":894},"# config.yaml 写 model_list \u002F fallbacks \u002F litellm_settings\n",[826,903,905],{"class":828,"line":904},7,[826,906,889],{"emptyLinePlaceholder":560},[826,908,910,913,916,919],{"class":828,"line":909},8,[826,911,912],{"class":832},"docker",[826,914,915],{"class":836}," compose",[826,917,918],{"class":836}," up",[826,920,921],{"class":844}," -d\n",[826,923,925],{"class":828,"line":924},9,[826,926,889],{"emptyLinePlaceholder":560},[826,928,930],{"class":828,"line":929},10,[826,931,932],{"class":894},"# 生成虚拟 Key\n",[826,934,936,939,942,945,948],{"class":828,"line":935},11,[826,937,938],{"class":832},"curl",[826,940,941],{"class":844}," -X",[826,943,944],{"class":836}," POST",[826,946,947],{"class":836}," http:\u002F\u002Flocalhost:4000\u002Fkey\u002Fgenerate",[826,949,950],{"class":844}," \\\n",[826,952,954,957,960,963,966],{"class":828,"line":953},12,[826,955,956],{"class":844},"  -H",[826,958,959],{"class":836}," \"Authorization: Bearer ",[826,961,962],{"class":840},"$LITELLM_MASTER_KEY",[826,964,965],{"class":836},"\"",[826,967,950],{"class":844},[826,969,971,974],{"class":828,"line":970},13,[826,972,973],{"class":844},"  -d",[826,975,976],{"class":836}," '{\"models\":[\"gpt-5.4\",\"claude-sonnet-4.6\"],\"max_budget\":100}'\n",[826,978,980],{"class":828,"line":979},14,[826,981,889],{"emptyLinePlaceholder":560},[826,983,985],{"class":828,"line":984},15,[826,986,987],{"class":894},"# 业务方调用\n",[826,989,991,993,996],{"class":828,"line":990},16,[826,992,938],{"class":832},[826,994,995],{"class":836}," http:\u002F\u002Flocalhost:4000\u002Fv1\u002Fchat\u002Fcompletions",[826,997,950],{"class":844},[826,999,1001,1003,1006],{"class":828,"line":1000},17,[826,1002,956],{"class":844},[826,1004,1005],{"class":836}," \"Authorization: Bearer sk-xxx\"",[826,1007,950],{"class":844},[826,1009,1011,1013],{"class":828,"line":1010},18,[826,1012,973],{"class":844},[826,1014,1015],{"class":836}," '{\"model\":\"gpt-5.4\",\"messages\":[...]}'\n",[21,1017,249],{"id":249},[251,1019,1020,1034],{},[254,1021,1022],{},[257,1023,1024,1026,1028,1030,1032],{},[260,1025,262],{},[260,1027,267],{},[260,1029,11],{},[260,1031,270],{},[260,1033,273],{},[275,1035,1036,1049,1063,1075,1087,1100,1113,1129,1142],{},[257,1037,1038,1040,1043,1045,1047],{},[280,1039,282],{},[280,1041,1042],{},"OSS + Enterprise",[280,1044,285],{},[280,1046,291],{},[280,1048,285],{},[257,1050,1051,1054,1057,1059,1061],{},[280,1052,1053],{},"协议",[280,1055,1056],{},"MIT",[280,1058,335],{},[280,1060,1056],{},[280,1062,335],{},[257,1064,1065,1067,1069,1071,1073],{},[280,1066,315],{},[280,1068,321],{},[280,1070,318],{},[280,1072,321],{},[280,1074,143],{},[257,1076,1077,1079,1081,1083,1085],{},[280,1078,298],{},[280,1080,304],{},[280,1082,301],{},[280,1084,307],{},[280,1086,310],{},[257,1088,1089,1092,1094,1096,1098],{},[280,1090,1091],{},"虚拟 Key + 预算",[280,1093,321],{},[280,1095,335],{},[280,1097,321],{},[280,1099,321],{},[257,1101,1102,1104,1107,1109,1111],{},[280,1103,344],{},[280,1105,1106],{},"✅ 三类",[280,1108,321],{},[280,1110,321],{},[280,1112,321],{},[257,1114,1115,1118,1121,1124,1127],{},[280,1116,1117],{},"admin UI",[280,1119,1120],{},"✅ 内置",[280,1122,1123],{},"dashboard",[280,1125,1126],{},"✅ 中文 UI",[280,1128,321],{},[257,1130,1131,1133,1135,1137,1140],{},[280,1132,330],{},[280,1134,318],{},[280,1136,318],{},[280,1138,1139],{},"✅ EPay 内置",[280,1141,318],{},[257,1143,1144,1147,1150,1152,1155],{},[280,1145,1146],{},"集成观测",[280,1148,1149],{},"Langfuse\u002FProm",[280,1151,1123],{},[280,1153,1154],{},"仪表盘",[280,1156,1157],{},"原生",[21,1159,387],{"id":387},[36,1161,1162,1170,1176,1182,1188,1198,1204,1218],{},[39,1163,1164,51,1167,1169],{},[42,1165,1166],{},"必固定版本号",[53,1168,796],{}," \u002F 滚动 tag 在 2026-03 供应链事件后已是禁忌；用带签名验证的具体版本（如 v1.85.0）",[39,1171,1172,1175],{},[42,1173,1174],{},"salt_key 不能轮换","：初始化前生成 + 加密备份",[39,1177,1178,1181],{},[42,1179,1180],{},"Postgres 备份","：所有虚拟 Key + 预算都在 DB，必须定期备份",[39,1183,1184,1187],{},[42,1185,1186],{},"fallback 链别堆超 3 个","：失败叠加延迟 + 计费混乱",[39,1189,1190,1193,1194,1197],{},[42,1191,1192],{},"drop_params 谨慎开","：开 ",[53,1195,1196],{},"drop_params: true"," 会静默丢不兼容字段，调试时容易 confused",[39,1199,1200,1203],{},[42,1201,1202],{},"PII guardrail 不是 0 延迟","：Presidio 调用增 50–100ms，敏感场景再用",[39,1205,1206,1209,1210,1213,1214,1217],{},[42,1207,1208],{},"monorepo + workspace 模型映射","：业务方调 ",[53,1211,1212],{},"gpt-4"," → config 映射到 ",[53,1215,1216],{},"openai\u002Fgpt-5.4-mini","，要规划稳定映射表",[39,1219,1220,1223],{},[42,1221,1222],{},"国内自托管","：BYOK Key 走海外 API 仍然受网络影响，需要部署在能直连厂商的节点",[21,1225,429],{"id":428},[36,1227,1228,1231,1234,1237,1240,1243,1246,1249],{},[39,1229,1230],{},"✅ 中大型团队 \u002F 微服务架构",[39,1232,1233],{},"✅ 合规 \u002F 数据主权要求",[39,1235,1236],{},"✅ 多团队预算治理",[39,1238,1239],{},"✅ 想叠加 Langfuse \u002F Helicone \u002F Prometheus 观测",[39,1241,1242],{},"✅ BYOK 模式跨多厂商",[39,1244,1245],{},"❌ 不想运维（走 OpenRouter \u002F Portkey）",[39,1247,1248],{},"❌ 国内业务支付 + 中文运营（走 one-api \u002F new-api）",[39,1250,1251],{},"❌ 纯个人项目（SDK 足够，不需要 Proxy）",[21,1253,461],{"id":461},[36,1255,1256,1261,1265,1269],{},[39,1257,1258],{},[467,1259,1260],{"href":561},"OpenRouter 评测",[39,1262,1263],{},[467,1264,476],{"href":475},[39,1266,1267],{},[467,1268,482],{"href":481},[39,1270,1271],{},[467,1272,488],{"href":487},[21,1274,491],{"id":491},[215,1276,1277,1284,1291,1298],{},[39,1278,1279,1280],{},"NerdLevelTech — LiteLLM Proxy Production Tutorial 2026（含 v1.85.0 + 供应链事件细节）",[467,1281,1282],{"href":1282,"rel":1283},"https:\u002F\u002Fnerdleveltech.com\u002Flitellm-proxy-production-llm-gateway-tutorial",[501],[39,1285,1286,1287],{},"LiteLLM 官方文档 — Fallbacks \u002F Retries \u002F Cooldowns ",[467,1288,1289],{"href":1289,"rel":1290},"https:\u002F\u002Fdocs.litellm.ai\u002Fdocs\u002Fproxy\u002Freliability",[501],[39,1292,1293,1294],{},"Youngju.dev — LiteLLM Complete Guide 2026（SDK vs Proxy \u002F cost-based routing）",[467,1295,1296],{"href":1296,"rel":1297},"https:\u002F\u002Fwww.youngju.dev\u002Fblog\u002Fculture\u002F2026-03-25-litellm-unified-llm-api-proxy-guide-2025.en",[501],[39,1299,1300,1301],{},"GitHub — BerriAI\u002Flitellm README ",[467,1302,1303],{"href":1303,"rel":1304},"https:\u002F\u002Fgithub.com\u002FBerriAI\u002Flitellm",[501],[1306,1307,1308],"style",{},"html pre.shiki code .sScJk, html code.shiki .sScJk{--shiki-default:#6F42C1;--shiki-dark:#B392F0}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .sVt8B, html code.shiki .sVt8B{--shiki-default:#24292E;--shiki-dark:#E1E4E8}html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html pre.shiki code .szBVR, html code.shiki .szBVR{--shiki-default:#D73A49;--shiki-dark:#F97583}html pre.shiki code .sJ8bj, html code.shiki .sJ8bj{--shiki-default:#6A737D;--shiki-dark:#6A737D}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":524,"searchDepth":525,"depth":525,"links":1310},[1311,1312,1313,1314,1315,1316,1317,1318,1319,1320],{"id":23,"depth":528,"text":24},{"id":34,"depth":528,"text":34},{"id":118,"depth":528,"text":118},{"id":755,"depth":528,"text":756},{"id":815,"depth":528,"text":816},{"id":249,"depth":528,"text":249},{"id":387,"depth":528,"text":387},{"id":428,"depth":528,"text":429},{"id":461,"depth":528,"text":461},{"id":491,"depth":528,"text":491},"\u002Fimg\u002Ftools\u002Flitellm.webp","LiteLLM 真实评测：MIT 开源 LLM 网关，BerriAI 维护。SDK 模式（pip install litellm）+ Proxy 模式（Docker Compose 自托管）双形态，100+ 厂商统一 OpenAI 接口，虚拟 Key + 团队预算 + 三类 fallback + 成本追踪 + 内置 UI。2026-03 经历供应链事件后 v1.83+ 强化签名 + 镜像验证，生产请固定版本号。",[1324,1327,1330,1333],{"q":1325,"a":1326},"SDK 和 Proxy 模式区别？","SDK 模式（`from litellm import completion`）直接在 Python 里调，适合个人项目 \u002F 单服务。Proxy 模式启动一个 HTTP 服务（默认 4000 端口）+ Postgres，所有应用通过 OpenAI 兼容 endpoint 调用，支持虚拟 Key \u002F 团队 \u002F 预算 \u002F 配额 \u002F admin UI——团队 \u002F 多语言客户端 \u002F 生产建议走 Proxy。",{"q":1328,"a":1329},"Fallback 有几种？","三类：(1) general fallback——5xx\u002F429 错误切到备用模型；(2) content policy fallback——内容审查拒绝时切；(3) context window fallback——输入超 context 切大窗口模型（比如超 GPT-4 32k 自动切 Claude 200k）。可叠加 num_retries \u002F cooldown \u002F timeout 做完整 SRE 策略。",{"q":1331,"a":1332},"2026-03 供应链事件是什么？","2026-03-24 PyPI 上的 v1.82.7 \u002F v1.82.8 在被替换约 40 分钟内可能植入 Trivy CI token 外泄代码。BerriAI 在 v1.83+ 强化签名 + 镜像验证，生产部署务必固定到带签名验证的版本（如 v1.85.0），不要用 `:latest` tag。",{"q":1334,"a":1335},"和 OpenRouter 怎么选？","OpenRouter = SaaS（不用自己运维 + 直接付钱）；LiteLLM = OSS 自托管（自付服务器 + 完全控制数据 + BYOK 所有 Key）。中大型团队 \u002F 合规 \u002F 长期成本敏感 \u002F 自建 → LiteLLM Proxy；小团队 \u002F 不想运维 \u002F 快速迭代 → OpenRouter。",[557],{},[564,1339,912,1340],"proxy","kubernetes",[1342,1347],{"plan":1343,"price":1344,"features":1345,"notes":1346},"Open Source","$0（MIT）","SDK + Proxy + Postgres 虚拟 Key + 预算 + fallback + admin UI + 100+ providers","完全自托管，模型 \u002F 平台费走自付",{"plan":143,"price":579,"features":1348,"notes":1349},"SSO \u002F SAML + Audit + JWT auth + 高级路由 + SLA + on-prem 部署支持","联系 BerriAI 销售","MIT 开源免费 \u002F Enterprise SaaS 定制",[585],[588],{"power":590,"ux":591,"price":590,"cn_support":525,"stability":591},{"title":267,"description":1322},"LiteLLM 评测 2026：开源 LLM 网关，100+ 厂商统一 OpenAI 接口",[1357,1359,1361,1363],{"name":1358,"url":1282,"accessed":598},"NerdLevelTech — LiteLLM Proxy 生产教程 2026",{"name":1360,"url":1289,"accessed":598},"LiteLLM 官方文档 — Fallbacks",{"name":1362,"url":1296,"accessed":598},"Youngju.dev — LiteLLM Complete Guide 2026",{"name":1364,"url":1303,"accessed":598},"GitHub — BerriAI\u002Flitellm","tools\u002Fcoding\u002Fapi\u002Flitellm","MIT 开源 LLM 网关——SDK + Proxy 双形态，100+ 厂商统一 OpenAI 接口 + 虚拟 Key + 团队预算",[1368,1339,564,1369,1370,611,1371,613],"llm-gateway","opensource","mit","virtual-keys","想要『SaaS 体验 + 完全自托管』的开源 LLM 网关首选。组合『LiteLLM 网关 + Helicone\u002FLangfuse 观测』在中大型 dev org 是黄金组合。中文中转 + 业务支付走 one-api。","https:\u002F\u002Fwww.litellm.ai","lPcnAKqeWqMmusxJPPJHbUrv5ILkQuJsxwf_F338Mes",1785428432048]