[{"data":1,"prerenderedAt":173},["ShallowReactive",2],{"repo-weave-os-router-zh-CN":3,"i18n-weave-os-router-zh-CN":170},{"mirrorPool":4,"detail":13,"relatedProjects":60},[5,9],{"key":6,"label":7,"base":8,"type":6},"github","mirror.github","https:\u002F\u002Fgithub.com\u002F",{"key":10,"label":11,"base":12,"type":10},"auto-proxy","mirror.autoProxy","https:\u002F\u002Fgithub.xxlab.tech\u002F",{"owner":14,"repo":15,"repoId":16,"repoUrl":17,"starCount":18,"language":19,"description":20,"languages":21,"screenshots":29,"homepage":30,"pushedAt":31,"ownerAvatar":32,"ownerName":33,"ownerBio":34,"lastSyncTime":35,"lastReleaseTime":36,"tags":37,"license":45,"category":46,"subgroup":47,"releases":48},"weave-os","router",1222789989,"https:\u002F\u002Fgithub.com\u002Fweave-os\u002Frouter",5563,"Go","Model router for agentic systems. Routes every prompt to the right model in \u003C50ms. Cut costs 40-70% with just an endpoint change.",{"Go":22,"Shell":23,"TypeScript":24,"Python":25,"PLpgSQL":26,"Dockerfile":27,"Makefile":27,"JavaScript":27,"CSS":28,"HCL":28},84.5,5.7,4.5,3.4,1.5,0.1,0,[],"https:\u002F\u002Fweaveos.com\u002Fproducts\u002Frouter","2026-10-04","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F174961884?v=4","Weave","ML to quantify engineering","2026-10-04T13:14:33.928663+00:00","2026-07-13T00:44:31Z",[38,39,40,41,42,43,44],"agentic-coding","ai-gateway","anthropic","claude-code","codex","model-router","openai-compatible","Apache-2.0","ai","LLM推理\u002F微调工具",[49],{"tag":50,"publishAt":36,"semVer":51,"prerelease":52,"assets":53},"hmm-model-v1",[28,28,28],true,[54],{"fileName":55,"sizeMb":56,"sha256":57,"downloadCount":58,"originUrl":59},"hmm-router-frozen-v1.tar.gz",5.65,"b326bda644aac86831bae47d0b69de008046f62324a4b93a1eba702581c174df",3847,"https:\u002F\u002Fgithub.com\u002Fweave-os\u002Frouter\u002Freleases\u002Fdownload\u002Fhmm-model-v1\u002Fhmm-router-frozen-v1.tar.gz",[61,79,108,128,140,157],{"owner":62,"repo":63,"repoId":64,"category":46,"subgroup":47,"language":65,"tags":66,"starCount":68,"ownerUid":69,"latestTag":70,"platforms":71,"lastReleaseTime":76,"downloadCount":77,"description":78},"ggml-org","llama.cpp",612354784,"C++",[67],"ggml",131695,134263123,"b11734",[72,73,74,75],"windows","mac","linux","other","2026-10-19T14:08:40Z",481,"一个用C\u002FC++实现的大语言模型（LLM）推理引擎，支持多种模型格式，提供CLI、服务器和内置Web UI，可在本地运行LLM和VLM。",{"owner":80,"repo":81,"repoId":82,"category":46,"subgroup":47,"language":83,"tags":84,"starCount":101,"ownerUid":102,"latestTag":103,"platforms":104,"lastReleaseTime":105,"downloadCount":106,"description":107},"diegosouzapw","OmniRoute",1157102282,"TypeScript",[85,86,39,40,87,41,88,42,89,90,91,92,93,94,95,96,97,98,99,100],"a2a","ai-agents","claude","cline","copilot","cursor","deepseek","free-ai","gemini","kimi","llm-gateway","mcp","openai","openai-proxy","qwen","token-saver",72815,8016841,"v3.8.51",[72,73,74,75],"2026-09-30T01:41:50Z",4681,"一款免费的 MIT AI 网关，提供统一端点访问 290+ 提供商、500+ 模型，支持智能回退和令牌压缩。",{"owner":109,"repo":110,"repoId":111,"category":112,"subgroup":47,"language":113,"tags":114,"starCount":121,"ownerUid":122,"latestTag":123,"platforms":124,"lastReleaseTime":125,"downloadCount":126,"description":127},"AlexsJones","llmfit",1158588224,"tools","Rust",[115,116,117,118,119,120],"gguf","llm","localai","mlx","skill","unsloth",37743,1235925,"v1.1.17",[72,73,74],"2026-10-08T18:04:05Z",1447,"一个帮助用户根据硬件配置推荐最佳本地大语言模型的命令行工具，支持多GPU、MoE架构和多种量化格式。",{"owner":129,"repo":130,"repoId":131,"category":46,"subgroup":47,"language":83,"tags":132,"starCount":133,"ownerUid":134,"latestTag":135,"platforms":136,"lastReleaseTime":137,"downloadCount":138,"description":139},"tashfeenahmed","freellmapi",1217040185,[],37411,9307356,"v0.16.5",[72,73,74,75],"2026-10-19T11:30:09Z",1332,"一个聚合了34个免费LLM提供商、635个免费模型端点的统一API网关，每月提供约74亿免费令牌，支持智能路由、自动故障转移和加密密钥存储。",{"owner":141,"repo":142,"repoId":143,"category":46,"subgroup":47,"language":144,"tags":145,"starCount":150,"ownerUid":151,"latestTag":152,"platforms":153,"lastReleaseTime":154,"downloadCount":155,"description":156},"jundot","omlx",1157171418,"Python",[146,147,116,148,118,149],"apple-silicon","inference-server","macos","openai-api",22858,64250138,"v0.7.1rc1",[73,75],"2026-10-15T18:23:10Z",2195,"一个专为Apple Silicon优化的LLM推理服务器，支持连续批处理和SSD缓存，通过菜单栏进行管理。",{"owner":158,"repo":159,"repoId":160,"category":46,"subgroup":47,"language":19,"tags":161,"starCount":163,"ownerUid":164,"latestTag":165,"platforms":166,"lastReleaseTime":167,"downloadCount":168,"description":169},"yetone","magpie",1383809012,[41,42,91,162,116,148],"gemini-cli",16805,1206493,"v0.1.1293",[72,73],"2026-10-19T12:31:24Z",309,"Magpie 是一个本地运行的 AI 网关，让 Claude Code、Codex、Gemini CLI、OpenCode 等编程代理统一使用 Kimi、DeepSeek、GLM、ChatGPT 等模型，菜单栏一键切换，配额用尽自动切换下一账号。",{"description":171,"readmeHtml":172},"一个为代理系统设计的模型路由器，能在50ms内将每个提示词路由到最合适的模型，通过简单更换端点即可降低40-70%的成本，支持Anthropic、OpenAI、Gemini及开源模型。","\u003Ch2 id=\"项目概述\">项目概述\u003C\u002Fh2>\u003Cp>WorkWeave Router 是一个为代理系统设计的模型路由器，能在50ms内将每个提示词路由到最合适的模型。它作为一个即插即用的代理，支持 Anthropic Messages、OpenAI Chat Completions 和 Gemini 原生 API，通过一个小型本地嵌入器为每个请求选择最佳模型。\u003C\u002Fp>\n\u003Ch2 id=\"主要功能\">主要功能\u003C\u002Fh2>\u003Cul>\n\u003Cli>\u003Cstrong>快速路由\u003C\u002Fstrong>：每个请求在50ms内完成路由\u003C\u002Fli>\n\u003Cli>\u003Cstrong>多API支持\u003C\u002Fstrong>：支持 Anthropic Messages、OpenAI Chat Completions 和 Gemini 原生 API\u003C\u002Fli>\n\u003Cli>\u003Cstrong>开源模型支持\u003C\u002Fstrong>：通过 OpenRouter 支持 DeepSeek、Kimi、GLM、Qwen、Llama、Mistral 等\u003C\u002Fli>\n\u003Cli>\u003Cstrong>自带密钥\u003C\u002Fstrong>：提供商密钥保留在本地，静态加密\u003C\u002Fli>\n\u003Cli>\u003Cstrong>可观测性\u003C\u002Fstrong>：开箱即用的 OTLP 追踪，支持 Weave 仪表盘、Honeycomb、Datadog、Grafana 等\u003C\u002Fli>\n\u003Cli>\u003Cstrong>流式、工具调用和视觉支持\u003C\u002Fstrong>：完整支持主流功能\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"适用人群\">适用人群\u003C\u002Fh2>\u003Cul>\n\u003Cli>使用 Claude Code、Codex、Cursor 等 AI 编码代理的开发者\u003C\u002Fli>\n\u003Cli>希望优化多模型调用成本的团队\u003C\u002Fli>\n\u003Cli>需要统一 API 接入多种 LLM 的工程团队\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"快速上手\">快速上手\u003C\u002Fh2>\u003Col>\n\u003Cli>\u003Cstrong>最快方式\u003C\u002Fstrong>：通过 npx 一键安装，为 Claude Code、Codex、opencode 或 pi 配置路由器\u003Cdiv class=\"code-block\">\u003Cpre>\u003Ccode class=\"hljs language-bash\">npx @workweave\u002Frouter\u003C\u002Fcode>\u003C\u002Fpre>\u003C\u002Fdiv>安装程序会询问使用哪个工具、配置范围，并自动生成配置文件\u003C\u002Fli>\n\u003Cli>\u003Cstrong>指定工具安装\u003C\u002Fstrong>：\u003Cul>\n\u003Cli>Claude Code：\u003Ccode>npx @workweave\u002Frouter --claude\u003C\u002Fcode>\u003C\u002Fli>\n\u003Cli>Codex CLI：\u003Ccode>npx @workweave\u002Frouter --codex\u003C\u002Fcode>\u003C\u002Fli>\n\u003Cli>opencode：\u003Ccode>npx @workweave\u002Frouter --opencode\u003C\u002Fcode>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003C\u002Fli>\n\u003Cli>\u003Cstrong>自托管完整服务\u003C\u002Fstrong>：\u003Cdiv class=\"code-block\">\u003Cpre>\u003Ccode class=\"hljs language-bash\">echo \"OPENROUTER_API_KEY=sk-or-v1-...\" >> .env.local\nmake full-setup\u003C\u002Fcode>\u003C\u002Fpre>\u003C\u002Fdiv>路由器在 \u003Ccode>http:\u002F\u002Flocalhost:8080\u003C\u002Fcode> 运行，仪表盘在 \u003Ccode>http:\u002F\u002Flocalhost:8080\u002Fui\u002Fdashboard\u003C\u002Fcode>\u003C\u002Fli>\n\u003Cli>需要 Node ≥ 18\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Ch2 id=\"核心优势\">核心优势\u003C\u002Fh2>\u003Cul>\n\u003Cli>降低 40-70% 的 LLM 调用成本\u003C\u002Fli>\n\u003Cli>每个请求在 50ms 内完成路由，延迟极低\u003C\u002Fli>\n\u003Cli>统一的 API 接入，支持主流模型提供商和开源模型\u003C\u002Fli>\n\u003Cli>开箱即用的可观测性，支持 OTLP 追踪\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"系统要求\">系统要求\u003C\u002Fh2>\u003Cul>\n\u003Cli>操作系统：跨平台（macOS、Linux、Windows）\u003C\u002Fli>\n\u003Cli>额外依赖：Node.js ≥ 18（部分路径需要 \u003Ccode>jq\u003C\u002Fcode>）\u003C\u002Fli>\n\u003C\u002Ful>\n",1792450993321]