[{"data":1,"prerenderedAt":3396},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"review-openhuman-deep-review":9,"review-related-openhuman-deep-review":1338},{"tools":4,"reviews":5},78,26,{"tools":4,"reviews":5,"playbooks":7,"news":8},22,21,{"id":10,"title":11,"body":12,"cover":1318,"description":1319,"extension":1320,"lastVerified":1321,"meta":1322,"navigation":1323,"path":1324,"published":1325,"relatedTools":1326,"seo":1330,"stem":1331,"tags":1332,"updated":1325,"verdict":1336,"__hash__":1337},"review\u002Freview\u002Fopenhuman-deep-review.md","OpenHuman 深度评测：开源个人 AI 超级智能助手，真正懂你的数字分身",{"type":13,"value":14,"toc":1262},"minimark",[15,19,32,39,49,53,60,66,72,78,84,88,91,95,140,143,248,251,254,257,261,264,342,346,349,352,355,358,362,365,470,473,476,479,482,486,489,492,525,528,531,537,543,549,555,558,561,564,567,754,757,760,763,766,792,795,800,833,838,849,858,861,887,890,894,897,901,904,908,911,915,918,921,925,928,932,935,939,942,946,949,952,955,993,996,1056,1062,1065,1072,1075,1101,1104,1108,1112,1115,1119,1122,1126,1129,1133,1136,1140,1143,1147,1154,1164,1170,1176,1183,1186,1212,1215,1258],[16,17,18],"h2",{"id":18},"一句话结论",[20,21,22,23,27,28,31],"p",{},"如果你是",[24,25,26],"strong",{},"知识工作者 \u002F Obsidian 用户 \u002F 多工具协作场景 \u002F 关注数据隐私","，想要一个真正了解你、记得你、能自动学习的个人 AI 助手——",[24,29,30],{},"OpenHuman 在 2026 年是开源个人 AI 领域最具野心的选择之一","。Memory Tree 长期记忆系统 + 118+ 第三方 OAuth 集成 + TokenJuice 智能压缩 + 桌面原生 UI，7,800+ GitHub stars 印证了社区对其方向的认可。",[20,33,34,35,38],{},"但它",[24,36,37],{},"不是 ChatGPT 或 Claude 的替代品","：它定位的是\"个人数字分身\"而非\"通用对话 AI\"。Early Beta 阶段功能迭代快、文档尚不完善、中文支持依赖底层模型。如果你追求开箱即用、稳定成熟的体验，建议等正式版或选商业产品。",[40,41,46],"div",{"className":42},[43,44,45],"card","p-5","my-4",[20,47,48],{},"推荐建议：先用自托管模式（GPL-3.0 免费）体验 Memory Tree 和核心集成能力，确认满足需求后再考虑是否订阅付费版。初始同步需要 5-10 分钟，建议连接 3-5 个核心服务（Gmail + GitHub + Calendar）开始。",[16,50,52],{"id":51},"openhuman-真正在解决的问题","OpenHuman 真正在解决的问题",[20,54,55,56,59],{},"社区讨论\"为什么 OpenHuman 火\"经常聚焦在\"7.8k Stars\"、\"桌面吉祥物\"、\"Rust 架构\"。但深一层看，OpenHuman 是在解决",[24,57,58],{},"个人 AI 助手领域的四个核心痛点","：",[20,61,62,65],{},[24,63,64],{},"第一个痛点：AI 没有长期记忆。"," 每一次对话，AI 都像第一次见到你——它不知道你的背景、偏好、项目进展、历史决策。OpenAI 的标语直击要害：\"Every model in the world shares the same fundamental limitation: they are stateless.\" OpenHuman 的 Memory Tree 系统让 AI 拥有了持续更新的长期记忆，容量可达 10 亿 token。",[20,67,68,71],{},[24,69,70],{},"第二个痛点：数据孤岛。"," 现代人的数字生活分散在 Gmail、Slack、GitHub、Notion、Google Calendar 等十几个平台。传统 AI 助手只能获取单一对话上下文，无法跨平台理解你的完整工作流。OpenHuman 通过 118+ OAuth 集成，一键打通所有数据源，每 20 分钟自动同步最新信息。",[20,73,74,77],{},[24,75,76],{},"第三个痛点：高昂的上下文成本。"," Agent 系统最大的隐性成本不是模型调用，而是上下文注入。原始数据包含大量冗余——HTML 标签、长 URL、广告格式符号。OpenHuman 的 TokenJuice 智能压缩层最高可降低 80% 的 token 消耗，让 API 成本从\"烧钱\"变成\"可接受\"。",[20,79,80,83],{},[24,81,82],{},"第四个痛点：AI 工具的学习门槛。"," 大多数 AI Agent 需要终端操作、配置 API Key、编写 Prompt。OpenHuman 从设计之初就是桌面应用优先——图形界面、OAuth 一键授权、安装即用，不需要用户懂命令行。",[16,85,87],{"id":86},"memory-tree最核心的差异化能力","Memory Tree：最核心的差异化能力",[20,89,90],{},"Memory Tree 是 OpenHuman 的灵魂能力，也是它与所有竞品最根本的区别。",[92,93,94],"h3",{"id":94},"工作流程",[96,97,98,105,111,117,123,134],"ol",{},[99,100,101,104],"li",{},[24,102,103],{},"数据采集","：通过 118+ OAuth 连接自动拉取邮件、文档、聊天记录、代码仓库、日历事件等",[99,106,107,110],{},[24,108,109],{},"标准化处理","：所有数据转化为不超过 3000 token 的 canonical Markdown 块",[99,112,113,116],{},[24,114,115],{},"智能评分","：对每个信息块进行重要性评分（基于时效性、相关性、来源权重等维度）",[99,118,119,122],{},[24,120,121],{},"层级摘要","：折叠成 per-source \u002F per-topic \u002F per-day 的层级摘要树",[99,124,125,128,129,133],{},[24,126,127],{},"本地存储","：存入本地 SQLite 数据库，同时生成 ",[130,131,132],"code",{},".md"," 文件同步到 Obsidian 兼容的知识库",[99,135,136,139],{},[24,137,138],{},"持续更新","：每 20 分钟自动遍历所有活跃连接，增量更新记忆树",[92,141,142],{"id":142},"与传统记忆方案的区别",[144,145,146,165],"table",{},[147,148,149],"thead",{},[150,151,152,156,159,162],"tr",{},[153,154,155],"th",{},"维度",[153,157,158],{},"传统 Context Window",[153,160,161],{},"RAG 方案",[153,163,164],{},"OpenHuman Memory Tree",[166,167,168,185,201,217,232],"tbody",{},[150,169,170,174,177,180],{},[171,172,173],"td",{},"记忆容量",[171,175,176],{},"几万 token",[171,178,179],{},"百万级",[171,181,182],{},[24,183,184],{},"10 亿 token",[150,186,187,190,193,196],{},[171,188,189],{},"记忆结构",[171,191,192],{},"线性文本",[171,194,195],{},"向量检索",[171,197,198],{},[24,199,200],{},"层级摘要树",[150,202,203,206,209,212],{},[171,204,205],{},"更新频率",[171,207,208],{},"每次对话",[171,210,211],{},"手动索引",[171,213,214],{},[24,215,216],{},"自动 20 分钟",[150,218,219,222,225,227],{},[171,220,221],{},"可读性",[171,223,224],{},"不可读",[171,226,224],{},[171,228,229],{},[24,230,231],{},"Markdown 文件，Obsidian 可编辑",[150,233,234,237,240,243],{},[171,235,236],{},"隐私",[171,238,239],{},"云端",[171,241,242],{},"可控",[171,244,245],{},[24,246,247],{},"本地优先",[92,249,250],{"id":250},"实际体验",[20,252,253],{},"Memory Tree 最直观的感受是：AI 不需要你重复介绍自己。你问\"帮我看看下周的会议安排\"，它已经知道你的日历数据。你问\"上周那个项目的 PR 合并了吗\"，它已经同步了 GitHub 的更新。你问\"我和张三的邮件往来里提到的那份合同\"，它能从邮件历史中检索到。",[20,255,256],{},"这种\"它知道你是谁\"的体验，和传统 AI 的\"每次对话都是陌生人\"形成了鲜明对比。",[16,258,260],{"id":259},"tokenjuice被低估的成本杀手","TokenJuice：被低估的成本杀手",[20,262,263],{},"TokenJuice 是 OpenHuman 的智能压缩层，在数据进入 LLM 之前进行预处理。它的压缩手段包括：",[144,265,266,279],{},[147,267,268],{},[150,269,270,273,276],{},[153,271,272],{},"压缩手段",[153,274,275],{},"说明",[153,277,278],{},"效果",[166,280,281,292,303,314,325],{},[150,282,283,286,289],{},[171,284,285],{},"HTML → Markdown",[171,287,288],{},"去除 HTML 标签，保留纯文本内容",[171,290,291],{},"省 40-60%",[150,293,294,297,300],{},[171,295,296],{},"长 URL 缩短",[171,298,299],{},"用短标识符替代长 URL",[171,301,302],{},"省 5-10%",[150,304,305,308,311],{},[171,306,307],{},"非 ASCII 字符清理",[171,309,310],{},"去除无关格式符号",[171,312,313],{},"省 5-15%",[150,315,316,319,322],{},[171,317,318],{},"重复内容去重",[171,320,321],{},"识别并去除重复信息块",[171,323,324],{},"省 10-30%",[150,326,327,332,337],{},[171,328,329],{},[24,330,331],{},"综合",[171,333,334],{},[24,335,336],{},"全栈压缩",[171,338,339],{},[24,340,341],{},"最高省 80%",[92,343,345],{"id":344},"这意味着什么","这意味着什么？",[20,347,348],{},"假设你让 AI 分析一封包含 HTML 格式的邮件（5KB HTML → 1.5KB 纯文本），再读取一个 GitHub issue 页面（10KB → 2KB Markdown），搜索 3 个相关文档（15KB → 3KB）。原始数据 30KB 经压缩后约 6KB——token 消耗从约 8,000 降到约 1,600。",[20,350,351],{},"按 GPT-4o 价格计算，一次深度分析的成本从 $0.04 降到 $0.008。如果每天做 50 次，月成本从 $60 降到 $12。",[92,353,354],{"id":354},"注意事项",[20,356,357],{},"TokenJuice 的压缩并不完美。处理合同条款、财务报表、合规材料等对文字精确性要求极高的内容时，需要关注压缩是否丢失了关键细节。项目文档也明确提示了这一点。",[16,359,361],{"id":360},"_118-第三方集成一键接入你的数字生活","118+ 第三方集成：一键接入你的数字生活",[20,363,364],{},"OpenHuman 支持 118 个以上第三方服务的 OAuth 一键接入，覆盖范围广泛：",[144,366,367,380],{},[147,368,369],{},[150,370,371,374,377],{},[153,372,373],{},"类别",[153,375,376],{},"代表服务",[153,378,379],{},"用途",[166,381,382,393,404,415,426,437,448,459],{},[150,383,384,387,390],{},[171,385,386],{},"邮件",[171,388,389],{},"Gmail、Outlook",[171,391,392],{},"阅读邮件、检索历史通信",[150,394,395,398,401],{},[171,396,397],{},"文档",[171,399,400],{},"Notion、Google Docs、Obsidian",[171,402,403],{},"知识库同步、文档摘要",[150,405,406,409,412],{},[171,407,408],{},"代码",[171,410,411],{},"GitHub、GitLab",[171,413,414],{},"PR 状态、代码变更、Issue 追踪",[150,416,417,420,423],{},[171,418,419],{},"沟通",[171,421,422],{},"Slack、Discord",[171,424,425],{},"聊天记录检索、通知摘要",[150,427,428,431,434],{},[171,429,430],{},"日历",[171,432,433],{},"Google Calendar、Outlook Calendar",[171,435,436],{},"日程安排、会议提醒",[150,438,439,442,445],{},[171,440,441],{},"存储",[171,443,444],{},"Google Drive、Dropbox",[171,446,447],{},"文件检索、内容分析",[150,449,450,453,456],{},[171,451,452],{},"支付",[171,454,455],{},"Stripe",[171,457,458],{},"交易记录、账单查询",[150,460,461,464,467],{},[171,462,463],{},"项目管理",[171,465,466],{},"Linear、Jira",[171,468,469],{},"任务状态、Sprint 回顾",[92,471,472],{"id":472},"自动同步机制",[20,474,475],{},"每 20 分钟自动遍历所有活跃连接，拉取增量数据到本地记忆库。不需要手动\"告诉\"AI 今天收到了什么邮件、明天有什么会议——它自己就会知道。",[92,477,478],{"id":478},"实际体验中的问题",[20,480,481],{},"部分 OAuth 连接器的稳定性参差不齐。Gmail 和 GitHub 表现稳定，但部分小众服务的 token 会偶尔过期需要重新授权。自动同步频率固定 20 分钟，无法手动触发即时同步——你在紧急情况下不能\"立刻\"让 AI 同步最新邮件。",[16,483,485],{"id":484},"桌面原生体验ui-优先的设计哲学","桌面原生体验：UI 优先的设计哲学",[20,487,488],{},"与 OpenClaw、Hermes Agent 等终端优先的 Agent 不同，OpenHuman 从设计之初就是桌面应用优先。",[92,490,491],{"id":491},"技术架构",[493,494,495,501,507,513,519],"ul",{},[99,496,497,500],{},[24,498,499],{},"桌面框架","：Tauri（Rust 后端 + Web 前端）",[99,502,503,506],{},[24,504,505],{},"前端","：TypeScript \u002F JavaScript",[99,508,509,512],{},[24,510,511],{},"后端核心","：Rust",[99,514,515,518],{},[24,516,517],{},"包管理","：pnpm",[99,520,521,524],{},[24,522,523],{},"构建工具","：CMake + Ninja",[20,526,527],{},"Rust + Tauri 的选择意味着：相比 Electron 方案，内存占用更低、启动速度更快、安装包更小。",[92,529,530],{"id":530},"关键交互设计",[20,532,533,536],{},[24,534,535],{},"桌面吉祥物（Mascot）","：一个有表情、会说话的桌面小伙伴，能响应环境变化。它不只是装饰——当有新的邮件摘要、日历提醒或系统通知时，吉祥物会主动提醒你。",[20,538,539,542],{},[24,540,541],{},"Google Meet 参会","：可以以真实参与者身份加入你的线上会议。这意味着 AI 可以\"参加\"会议、记录会议纪要、提取行动项——这是目前大多数 AI 助手做不到的。",[20,544,545,548],{},[24,546,547],{},"语音交互","：支持语音输入（STT）和语音输出（ElevenLabs TTS），带唇形同步。打字不方便时可以直接语音对话。",[20,550,551,554],{},[24,552,553],{},"后台持续思考","：即使你不主动交互，AI 也在后台持续分析和整理信息——总结未读邮件、提取关键文档内容、更新记忆树结构。",[92,556,557],{"id":557},"资源占用",[20,559,560],{},"桌面吉祥物 + 后台同步 + 持续思考，对资源有一定要求。建议 8GB+ 内存。低配机器（4GB 内存）上吉祥物动画会略卡顿，可以关闭吉祥物功能以节省资源。",[16,562,563],{"id":563},"与同类产品对比",[20,565,566],{},"OpenHuman 在 GitHub README 和相关社区文章中提供了与主流产品的对比：",[144,568,569,590],{},[147,570,571],{},[150,572,573,576,579,582,585],{},[153,574,575],{},"特性",[153,577,578],{},"Claude Cowork",[153,580,581],{},"OpenClaw",[153,583,584],{},"Hermes Agent",[153,586,587],{},[24,588,589],{},"OpenHuman",[166,591,592,608,624,640,659,677,692,708,723,736],{},[150,593,594,597,600,603,605],{},[171,595,596],{},"开源",[171,598,599],{},"❌ 专有",[171,601,602],{},"✅ MIT",[171,604,602],{},[171,606,607],{},"✅ GPL-3.0",[150,609,610,613,616,619,621],{},[171,611,612],{},"上手难度",[171,614,615],{},"✅ 桌面+CLI",[171,617,618],{},"⚠️ 终端优先",[171,620,618],{},[171,622,623],{},"✅ UI 优先，几分钟上手",[150,625,626,629,632,635,637],{},[171,627,628],{},"成本",[171,630,631],{},"⚠️ 订阅+附加",[171,633,634],{},"⚠️ 自备模型",[171,636,634],{},[171,638,639],{},"✅ 一价全包+TokenJuice",[150,641,642,645,648,651,654],{},[171,643,644],{},"长期记忆",[171,646,647],{},"⚠️ 对话级",[171,649,650],{},"⚠️ 依赖插件",[171,652,653],{},"✅ 自学习",[171,655,656],{},[24,657,658],{},"Memory Tree+Obsidian",[150,660,661,664,667,670,672],{},[171,662,663],{},"集成数量",[171,665,666],{},"⚠️ 少",[171,668,669],{},"⚠️ 需自建",[171,671,669],{},[171,673,674],{},[24,675,676],{},"118+ OAuth 集成",[150,678,679,682,685,687,689],{},[171,680,681],{},"自动同步",[171,683,684],{},"❌ 无",[171,686,684],{},[171,688,684],{},[171,690,691],{},"✅ 20 分钟自动抓取",[150,693,694,697,700,703,705],{},[171,695,696],{},"模型路由",[171,698,699],{},"❌ 单一",[171,701,702],{},"⚠️ 手动",[171,704,702],{},[171,706,707],{},"✅ 内置智能路由",[150,709,710,713,716,718,720],{},[171,711,712],{},"桌面吉祥物",[171,714,715],{},"❌",[171,717,715],{},[171,719,715],{},[171,721,722],{},"✅",[150,724,725,727,729,732,734],{},[171,726,236],{},[171,728,239],{},[171,730,731],{},"本地",[171,733,731],{},[171,735,247],{},[150,737,738,741,744,747,749],{},[171,739,740],{},"技术栈",[171,742,743],{},"专有",[171,745,746],{},"Python",[171,748,746],{},[171,750,751],{},[24,752,753],{},"Rust (Tauri)",[92,755,756],{"id":756},"一句话总结",[20,758,759],{},"OpenHuman 的优势在于**\"开箱即用的个人记忆\"**——别人需要插件、配置、自建连接器的地方，它已经内置好了。Claude Cowork 和 OpenClaw 更偏向通用任务执行，OpenHuman 更偏向\"了解你这个人\"。",[16,761,762],{"id":762},"安装与上手",[92,764,765],{"id":765},"系统要求",[493,767,768,774,780,786],{},[99,769,770,773],{},[24,771,772],{},"操作系统","：Windows \u002F macOS \u002F Linux",[99,775,776,779],{},[24,777,778],{},"运行时","：Node.js 24+（开发）、pnpm 10.10.0（开发）、Rust 1.93.0（开发）",[99,781,782,785],{},[24,783,784],{},"内存","：建议 8GB+",[99,787,788,791],{},[24,789,790],{},"网络","：需要稳定网络连接（OAuth 授权 + 模型调用）",[92,793,794],{"id":794},"安装方式",[20,796,797],{},[24,798,799],{},"macOS \u002F Linux：",[801,802,807],"pre",{"className":803,"code":804,"language":805,"meta":806,"style":806},"language-bash shiki shiki-themes github-light github-dark","curl -fsSL https:\u002F\u002Fraw.githubusercontent.com\u002Ftinyhumansai\u002Fopenhuman\u002Fmain\u002Fscripts\u002Finstall.sh | bash\n","bash","",[130,808,809],{"__ignoreMap":806},[810,811,814,818,822,826,830],"span",{"class":812,"line":813},"line",1,[810,815,817],{"class":816},"sScJk","curl",[810,819,821],{"class":820},"sj4cs"," -fsSL",[810,823,825],{"class":824},"sZZnC"," https:\u002F\u002Fraw.githubusercontent.com\u002Ftinyhumansai\u002Fopenhuman\u002Fmain\u002Fscripts\u002Finstall.sh",[810,827,829],{"class":828},"szBVR"," |",[810,831,832],{"class":816}," bash\n",[20,834,835],{},[24,836,837],{},"Windows（PowerShell）：",[801,839,843],{"className":840,"code":841,"language":842,"meta":806,"style":806},"language-powershell shiki shiki-themes github-light github-dark","irm https:\u002F\u002Fraw.githubusercontent.com\u002Ftinyhumansai\u002Fopenhuman\u002Fmain\u002Fscripts\u002Finstall.ps1 | iex\n","powershell",[130,844,845],{"__ignoreMap":806},[810,846,847],{"class":812,"line":813},[810,848,841],{},[20,850,851,852],{},"也可以直接从官网下载安装包：",[853,854,855],"a",{"href":855,"rel":856},"https:\u002F\u002Ftinyhumans.ai\u002Fopenhuman",[857],"nofollow",[92,859,860],{"id":860},"首次使用流程",[96,862,863,869,875,881],{},[99,864,865,868],{},[24,866,867],{},"安装并启动","：下载安装包或运行一键安装脚本",[99,870,871,874],{},[24,872,873],{},"OAuth 授权","：连接你的核心服务（建议至少 3-5 个：邮箱 + 日历 + 代码仓库 + 文档）",[99,876,877,880],{},[24,878,879],{},"等待同步","：初始同步需要 5-10 分钟（取决于数据量），Memory Tree 自动构建",[99,882,883,886],{},[24,884,885],{},"开始对话","：AI 已经知道你的工作背景——可以问\"我今天有什么会议？\"或\"帮我总结上周的邮件\"",[16,888,889],{"id":889},"适用场景",[92,891,893],{"id":892},"_1-个人知识管理","1. 个人知识管理",[20,895,896],{},"你是 Obsidian 用户，日常管理大量笔记和文档。OpenHuman 自动将邮箱、文档、聊天记录转化为 Markdown 文件同步到 Obsidian 知识库。你的知识库不再是\"手动维护\"的，而是\"AI 自动补充\"的。",[92,898,900],{"id":899},"_2-多工具协作场景","2. 多工具协作场景",[20,902,903],{},"你的工作流跨越 Gmail、Slack、GitHub、Notion、Google Calendar 等多个平台。传统方式需要手动在工具间切换，信息割裂。OpenHuman 打通所有数据源，让你在一个地方了解全貌。",[92,905,907],{"id":906},"_3-隐私敏感场景","3. 隐私敏感场景",[20,909,910],{},"你不想把个人数据上传到商业 AI 产品（ChatGPT、Claude 等）的云端服务器。OpenHuman 核心记忆数据存储在本地 SQLite，可以通过 Ollama 运行本地 LLM，敏感任务完全不上云。",[92,912,914],{"id":913},"_4-早期尝鲜者","4. 早期尝鲜者",[20,916,917],{},"你愿意尝试 Early Beta 产品，能接受偶尔的 bug 和功能迭代。OpenHuman 的方向正确、架构合理、社区活跃，值得持续关注。",[16,919,920],{"id":920},"不推荐场景",[92,922,924],{"id":923},"_1-追求极致轻量","1. 追求极致轻量",[20,926,927],{},"你只需要一个简单的 AI 聊天工具，不需要 118+ 集成、不需要长期记忆、不需要桌面吉祥物。OpenHuman 的设计是\"大而全\"的，对于轻量需求来说太重了。",[92,929,931],{"id":930},"_2-完全离线运行","2. 完全离线运行",[20,933,934],{},"虽然数据本地存储，但 OAuth 连接需要网络授权，远程模型调用需要网络。如果你希望完全离线，甚至不需要连接任何第三方服务，OpenHuman 当前不适合。",[92,936,938],{"id":937},"_3-生产级稳定","3. 生产级稳定",[20,940,941],{},"Early Beta 阶段，功能迭代快，偶尔有 breaking change。文档尚在完善中，部分功能缺少详细说明。如果你的工作不能接受偶发的不稳定，建议等正式版。",[92,943,945],{"id":944},"_4-中文优先场景","4. 中文优先场景",[20,947,948],{},"当前英文体验最佳，中文支持依赖底层模型（GPT-4o \u002F Claude 中文效果好，但本地模型的中文能力待验证）。如果你主要使用中文且需要深度本地化体验，建议观望。",[16,950,951],{"id":951},"价格与运行成本",[20,953,954],{},"OpenHuman 提供两种模式：",[144,956,957,969],{},[147,958,959],{},[150,960,961,964,967],{},[153,962,963],{},"模式",[153,965,966],{},"费用",[153,968,275],{},[166,970,971,982],{},[150,972,973,976,979],{},[171,974,975],{},"统一订阅",[171,977,978],{},"待定（Early Beta）",[171,980,981],{},"一价全包所有模型 + 自动路由",[150,983,984,987,990],{},[171,985,986],{},"自托管（GPL-3.0）",[171,988,989],{},"$0",[171,991,992],{},"自备 LLM API 或本地 Ollama 模型",[20,994,995],{},"自托管模式下，一次中等任务（5-10 步 + Memory Tree 检索）的 API 费用：",[144,997,998,1011],{},[147,999,1000],{},[150,1001,1002,1005,1008],{},[153,1003,1004],{},"模型",[153,1006,1007],{},"单次成本",[153,1009,1010],{},"月度估算（每天 10 次）",[166,1012,1013,1024,1035,1046],{},[150,1014,1015,1018,1021],{},[171,1016,1017],{},"GPT-4o",[171,1019,1020],{},"$0.05-0.20",[171,1022,1023],{},"$15-60",[150,1025,1026,1029,1032],{},[171,1027,1028],{},"GPT-4o-mini",[171,1030,1031],{},"$0.005-0.02",[171,1033,1034],{},"$1.5-6",[150,1036,1037,1040,1043],{},[171,1038,1039],{},"Claude Sonnet 4",[171,1041,1042],{},"$0.08-0.25",[171,1044,1045],{},"$24-75",[150,1047,1048,1051,1053],{},[171,1049,1050],{},"本地 Ollama 模型",[171,1052,989],{},[171,1054,1055],{},"$0（电费忽略）",[20,1057,1058,1061],{},[24,1059,1060],{},"省钱策略","：日常查询用 GPT-4o-mini，深度分析用 GPT-4o \u002F Claude。TokenJuice 压缩后，大部分日常任务的 token 消耗在 500-2,000 之间，成本极低。",[16,1063,1064],{"id":1064},"未来展望",[20,1066,1067,1068,1071],{},"OpenHuman 代表了 AI 产品的一个新方向：",[24,1069,1070],{},"从\"通用智能\"走向\"个人智能\"","。",[20,1073,1074],{},"当前阶段，它更像是一个\"技术 demo\"——展示了个人 AI 助手应该是什么样子。真正的挑战在于：",[96,1076,1077,1083,1089,1095],{},[99,1078,1079,1082],{},[24,1080,1081],{},"连接器生态的深度","：118+ 集成听起来很多，但每个连接器的质量、稳定性、API 覆盖度需要持续打磨",[99,1084,1085,1088],{},[24,1086,1087],{},"记忆系统的准确性","：摘要的质量、评分算法的合理性、检索的精确度，决定了记忆系统的实用价值",[99,1090,1091,1094],{},[24,1092,1093],{},"隐私边界的清晰度","：OAuth 连接意味着服务商可知你的连接状态，数据本地存储但模型调用可能涉及网络传输——这些边界需要更清晰地向用户说明",[99,1096,1097,1100],{},[24,1098,1099],{},"商业模式的可持续性","：统一订阅 + 开源自托管并行的模式，需要在营收和社区之间找到平衡",[20,1102,1103],{},"但方向是对的。如果你对个人 AI Agent 感兴趣，OpenHuman 绝对值得持续关注。",[16,1105,1107],{"id":1106},"faq","FAQ",[92,1109,1111],{"id":1110},"openhuman-和-chatgpt-claude-有什么不同","OpenHuman 和 ChatGPT \u002F Claude 有什么不同？",[20,1113,1114],{},"ChatGPT \u002F Claude 是通用对话 AI，每次对话从零开始，没有你的长期记忆。OpenHuman 通过 Memory Tree 系统持续学习你的邮件、日历、文档、代码仓库，构建一个持续更新的个人记忆库。它知道\"你是谁\"而非只是\"你问了什么\"。",[92,1116,1118],{"id":1117},"openhuman-的数据安全如何","OpenHuman 的数据安全如何？",[20,1120,1121],{},"核心记忆数据存储在本地 SQLite 数据库，不上传云端。数据在设备端加密存储。可通过 Ollama 运行本地 LLM 实现敏感任务完全离线。但 OAuth 连接和远程模型调用仍涉及网络传输。",[92,1123,1125],{"id":1124},"memory-tree-和-rag-有什么区别","Memory Tree 和 RAG 有什么区别？",[20,1127,1128],{},"RAG（检索增强生成）是在查询时从外部知识库检索相关文档片段，本质是\"搜 + 答\"。Memory Tree 是持续构建的层级摘要树，AI 拥有结构化的长期记忆——不仅知道\"有什么文档\"，还知道\"文档之间的关联\"、\"历史决策的上下文\"、\"信息的时效性权重\"。",[92,1130,1132],{"id":1131},"tokenjuice-会丢失信息吗","TokenJuice 会丢失信息吗？",[20,1134,1135],{},"对于大多数场景（邮件、文档、聊天记录），压缩后的 Markdown 保留了所有关键信息。但对于合同条款、财务报表、合规材料等对文字精确性要求极高的内容，建议关注压缩是否丢失了细节。项目文档也提示了这一点。",[92,1137,1139],{"id":1138},"支持移动端吗","支持移动端吗？",[20,1141,1142],{},"目前仅桌面端（Windows \u002F macOS \u002F Linux）。未来可能有移动端计划，但暂无明确时间表。",[16,1144,1146],{"id":1145},"aiho-推荐结论","AIHO 推荐结论",[20,1148,1149,1150,1153],{},"OpenHuman 是 2026 年开源个人 AI 助手领域",[24,1151,1152],{},"最具创新性的项目之一","——7.8k stars 不是白来的。Memory Tree + 118+ 集成 + TokenJuice 压缩 + 桌面原生 UI，能力栈体现了团队对\"个人 AI 助手\"的深入理解。Rust + Tauri 的技术选型保证了性能和用户体验。",[20,1155,34,1156,1159,1160,1163],{},[24,1157,1158],{},"不是适合所有人的通用 AI 工具","。它的核心价值在于\"了解你这个人\"——如果你需要的是通用对话、代码生成、写作辅助，ChatGPT \u002F Claude 仍然是更好的选择。OpenHuman 的价值在于",[24,1161,1162],{},"个人记忆和自动化","：它知道你的工作背景、自动同步你的数字生活、替你记住那些你记不住的事情。",[20,1165,1166,1169],{},[24,1167,1168],{},"选 OpenHuman 如果","：你是知识工作者，想要一个真正了解你的 AI 助手，愿意为长期记忆和自动集成付出上手成本。",[20,1171,1172,1175],{},[24,1173,1174],{},"别选 OpenHuman 如果","：要开箱即用（去 ChatGPT \u002F Claude）、要纯中文深度优化（去国产 AI 产品）、要生产级稳定（等正式版）、追求极致轻量（不需要这些集成功能）。",[20,1177,1178,1179,1182],{},"最好的用法是",[24,1180,1181],{},"组合使用","：ChatGPT \u002F Claude 做通用对话和创作，OpenHuman 做个人记忆和自动化助手。两者互补，共同构建你的 AI 工作流。",[16,1184,1185],{"id":1185},"相关阅读",[493,1187,1188,1194,1200,1206],{},[99,1189,1190],{},[853,1191,1193],{"href":1192},"\u002Fagent\u002Fgeneral\u002Fopenhuman.html","OpenHuman 工具卡：个人 AI 超级智能助手",[99,1195,1196],{},[853,1197,1199],{"href":1198},"\u002Freview\u002Fhermes-agent-deep-review.html","Hermes Agent 深度评测：自学习 AI Agent",[99,1201,1202],{},[853,1203,1205],{"href":1204},"\u002Freview\u002Fopenmanus-deep-review.html","OpenManus 深度评测：开源版 Manus",[99,1207,1208],{},[853,1209,1211],{"href":1210},"\u002Freview\u002Fmanus-deep-review.html","Manus 深度评测：通用 Agent 天花板",[16,1213,1214],{"id":1214},"来源",[96,1216,1217,1224,1230,1237,1244,1251],{},[99,1218,1219,1220],{},"OpenHuman GitHub 主仓库 + TinyHumans 组织 ",[853,1221,1222],{"href":1222,"rel":1223},"https:\u002F\u002Fgithub.com\u002Ftinyhumansai\u002Fopenhuman",[857],[99,1225,1226,1227],{},"OpenHuman 官网 ",[853,1228,855],{"href":855,"rel":1229},[857],[99,1231,1232,1233],{},"OpenHuman 文档 ",[853,1234,1235],{"href":1235,"rel":1236},"https:\u002F\u002Ftinyhumans.gitbook.io\u002Fopenhuman",[857],[99,1238,1239,1240],{},"OpenHuman Discord ",[853,1241,1242],{"href":1242,"rel":1243},"https:\u002F\u002Fdiscord.tinyhumans.ai\u002F",[857],[99,1245,1246,1247],{},"CSDN — OpenHuman 深度调研报告 ",[853,1248,1249],{"href":1249,"rel":1250},"https:\u002F\u002Fblog.csdn.net\u002FAiKnighterrant\u002Farticle\u002Fdetails\u002F161504465",[857],[99,1252,1253,1254],{},"CSDN — OpenHuman 自动了解用户的 AI Agent 新范式 ",[853,1255,1256],{"href":1256,"rel":1257},"https:\u002F\u002Fblog.csdn.net\u002Fshaobingj126\u002Farticle\u002Fdetails\u002F161169056",[857],[1259,1260,1261],"style",{},"html pre.shiki code .sScJk, html code.shiki .sScJk{--shiki-default:#6F42C1;--shiki-dark:#B392F0}html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .szBVR, html code.shiki .szBVR{--shiki-default:#D73A49;--shiki-dark:#F97583}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":806,"searchDepth":1263,"depth":1263,"links":1264},3,[1265,1267,1268,1273,1277,1281,1286,1289,1294,1300,1306,1307,1308,1315,1316,1317],{"id":18,"depth":1266,"text":18},2,{"id":51,"depth":1266,"text":52},{"id":86,"depth":1266,"text":87,"children":1269},[1270,1271,1272],{"id":94,"depth":1263,"text":94},{"id":142,"depth":1263,"text":142},{"id":250,"depth":1263,"text":250},{"id":259,"depth":1266,"text":260,"children":1274},[1275,1276],{"id":344,"depth":1263,"text":345},{"id":354,"depth":1263,"text":354},{"id":360,"depth":1266,"text":361,"children":1278},[1279,1280],{"id":472,"depth":1263,"text":472},{"id":478,"depth":1263,"text":478},{"id":484,"depth":1266,"text":485,"children":1282},[1283,1284,1285],{"id":491,"depth":1263,"text":491},{"id":530,"depth":1263,"text":530},{"id":557,"depth":1263,"text":557},{"id":563,"depth":1266,"text":563,"children":1287},[1288],{"id":756,"depth":1263,"text":756},{"id":762,"depth":1266,"text":762,"children":1290},[1291,1292,1293],{"id":765,"depth":1263,"text":765},{"id":794,"depth":1263,"text":794},{"id":860,"depth":1263,"text":860},{"id":889,"depth":1266,"text":889,"children":1295},[1296,1297,1298,1299],{"id":892,"depth":1263,"text":893},{"id":899,"depth":1263,"text":900},{"id":906,"depth":1263,"text":907},{"id":913,"depth":1263,"text":914},{"id":920,"depth":1266,"text":920,"children":1301},[1302,1303,1304,1305],{"id":923,"depth":1263,"text":924},{"id":930,"depth":1263,"text":931},{"id":937,"depth":1263,"text":938},{"id":944,"depth":1263,"text":945},{"id":951,"depth":1266,"text":951},{"id":1064,"depth":1266,"text":1064},{"id":1106,"depth":1266,"text":1107,"children":1309},[1310,1311,1312,1313,1314],{"id":1110,"depth":1263,"text":1111},{"id":1117,"depth":1263,"text":1118},{"id":1124,"depth":1263,"text":1125},{"id":1131,"depth":1263,"text":1132},{"id":1138,"depth":1263,"text":1139},{"id":1145,"depth":1266,"text":1146},{"id":1185,"depth":1266,"text":1185},{"id":1214,"depth":1266,"text":1214},"\u002Fog\u002Freview\u002Fopenhuman.svg","OpenHuman 深度评测：TinyHumans 团队 2026-05 开源的个人 AI 超级智能助手，7.8k+ GitHub stars。本文写它真正解决的问题——Memory Tree 长期记忆、118+ 服务集成、TokenJuice 智能压缩、桌面原生体验、与 Claude Cowork \u002F OpenClaw \u002F Hermes Agent 的对比、适用场景和 4 类不推荐场景。AIHO 编辑部基于官方文档与社区反馈整理。","md",null,{},true,"\u002Freview\u002Fopenhuman-deep-review","2026-07-17",[1327,1328,1329],"agent\u002Fgeneral\u002Fopenhuman","agent\u002Fgeneral\u002Fhermes-agent","agent\u002Fgeneral\u002Fopenmanus",{"title":11,"description":1319},"review\u002Fopenhuman-deep-review",[589,1333,1334,596,1335],"AI Agent","个人AI","深度评测","追求个人 AI 助手真正了解你、记得你的知识工作者首选。Memory Tree + 118+ 集成 + TokenJuice 压缩 + 桌面原生 UI，让 AI 从『通用聊天工具』升级为『你的数字分身』。但 Early Beta 阶段不稳定、中文支持待优化、生产场景需谨慎。","kQR3kqqLKxflZfvIfMhA4lXP-37Y3UPew9k15UMDnps",[1339,2108,2684],{"id":1340,"title":584,"alternatives":1341,"api_compatible":1344,"body":1348,"category":2032,"chinese_friendly":1266,"cover":2033,"description":2034,"domestic":2035,"extension":1320,"faq":1321,"free":2035,"github":2036,"languages":2037,"lastVerified":1321,"meta":2039,"models":2040,"navigation":1323,"notSuitable":2045,"opensource":1323,"path":2050,"pillar":2051,"platforms":2052,"priceTable":2056,"pricing":2068,"published":2069,"relatedPlaybooks":1321,"relatedReviews":2070,"score":2072,"self_host":1323,"seo":2073,"seoTitle":2074,"slug":1328,"sources":2075,"stem":2090,"suitable":2091,"tagline":2097,"tags":2098,"updated":2069,"verdict":2106,"website":2036,"__hash__":2107},"tools\u002Ftools\u002Fagent\u002Fgeneral\u002Fhermes-agent.md",[1342,1329,1343],"agent\u002Fgeneral\u002Fmanus","agent\u002Fdesktop\u002Fopenclaw",[1345,1346,1347],"openai","anthropic","local",{"type":13,"value":1349,"toc":2010},[1350,1352,1359,1365,1375,1379,1383,1389,1415,1422,1426,1433,1447,1450,1454,1460,1540,1546,1549,1552,1558,1576,1579,1582,1585,1605,1608,1611,1614,1628,1631,1634,1703,1718,1721,1727,1741,1744,1747,1773,1776,1829,1835,1838,1942,1944,1950,1956,1962,1968,1974,1976,2007],[16,1351,18],{"id":18},[20,1353,1354,1355,1358],{},"如果你想要一个",[24,1356,1357],{},"完全开源、能自我进化、有长期记忆、可以部署在十几个消息平台上的 AI Agent","——Hermes Agent 在 2026 年是最成熟的选择。GitHub 10 万 Star，2 月开源到 5 月就翻了一倍。",[20,1360,1361,1362],{},"但它的门槛比 Manus 高一个量级：没有漂亮的 Web 界面，需要自己装 Python 环境、配置 LLM API、管理记忆数据库。",[24,1363,1364],{},"它面向的是\"想拥有自己 Agent 基础设施\"的极客，不是\"想要一个好用 AI 助手\"的普通用户。",[1366,1367,1369],"callout",{"type":1368},"info",[20,1370,1371,1374],{},[24,1372,1373],{},"定位区分","：Manus 是\"帮你干完事的云端 Agent\"，OpenClaw 是\"常驻你电脑的 AI 操作系统\"，Hermes Agent 是\"你能完全掌控和改造的自进化 Agent\"。三者目标不同，选型看你需要的是\"用\"还是\"拥有\"。",[16,1376,1378],{"id":1377},"hermes-agent-解决的核心问题","Hermes Agent 解决的核心问题",[92,1380,1382],{"id":1381},"问题-1agent-没有长期记忆","问题 1：Agent 没有\"长期记忆\"",[20,1384,1385,1386,59],{},"大多数 AI Agent 的记忆只有当前会话的上下文窗口。你昨天让它做的事、偏好、上下文，今天全忘了。Hermes Agent 内置了",[24,1387,1388],{},"分层记忆系统",[493,1390,1391,1397,1403,1409],{},[99,1392,1393,1396],{},[24,1394,1395],{},"工作记忆","：当前会话上下文，类似人类短期记忆",[99,1398,1399,1402],{},[24,1400,1401],{},"情景记忆","：记录每次交互的时间、事件、结果，支持按时间线回溯",[99,1404,1405,1408],{},[24,1406,1407],{},"语义记忆","：从交互中抽取知识（你告诉它的偏好、事实、规则），长期保存",[99,1410,1411,1414],{},[24,1412,1413],{},"技能记忆","：Agent 自动总结\"怎么做某件事\"的步骤，下次直接调用",[20,1416,1417,1418,1421],{},"这意味着 Hermes Agent 是",[24,1419,1420],{},"越用越好用","的——它会记住你的工作习惯、项目上下文、你纠正过的错误。",[92,1423,1425],{"id":1424},"问题-2单个模型能力有天花板","问题 2：单个模型能力有天花板",[20,1427,1428,1429,1432],{},"2026 年 6 月，Nous Research 给 Hermes Agent 加了 ",[24,1430,1431],{},"MoA（Mixture of Agents）"," 功能：",[493,1434,1435,1438,1441,1444],{},[99,1436,1437],{},"多个 Agent 实例并行处理同一个任务",[99,1439,1440],{},"每个 Agent 可以用不同模型（Claude \u002F GPT \u002F Hermes \u002F 本地模型）",[99,1442,1443],{},"结果由一个\"聚合 Agent\"合并去重、交叉验证",[99,1445,1446],{},"最终输出质量 > 任何单个模型",[20,1448,1449],{},"实际效果：用 3 个中档模型做 MoA，输出质量可以接近 1 个顶级模型，但成本只有 1\u002F3。",[92,1451,1453],{"id":1452},"问题-3agent-只能在一个地方用","问题 3：Agent 只能在一个地方用",[20,1455,1456,1457,59],{},"Hermes Agent 支持 ",[24,1458,1459],{},"14 个消息渠道",[144,1461,1462,1472],{},[147,1463,1464],{},[150,1465,1466,1469],{},[153,1467,1468],{},"平台",[153,1470,1471],{},"状态",[166,1473,1474,1482,1489,1496,1503,1510,1517,1524,1532],{},[150,1475,1476,1479],{},[171,1477,1478],{},"Telegram",[171,1480,1481],{},"✅ 推荐，最稳定",[150,1483,1484,1487],{},[171,1485,1486],{},"Discord",[171,1488,722],{},[150,1490,1491,1494],{},[171,1492,1493],{},"Slack",[171,1495,722],{},[150,1497,1498,1501],{},[171,1499,1500],{},"WhatsApp",[171,1502,722],{},[150,1504,1505,1508],{},[171,1506,1507],{},"Signal",[171,1509,722],{},[150,1511,1512,1515],{},[171,1513,1514],{},"Email",[171,1516,722],{},[150,1518,1519,1522],{},[171,1520,1521],{},"CLI（终端）",[171,1523,722],{},[150,1525,1526,1529],{},[171,1527,1528],{},"Web UI",[171,1530,1531],{},"✅ 基础版",[150,1533,1534,1537],{},[171,1535,1536],{},"微信",[171,1538,1539],{},"⚠️ 非官方，不稳定",[20,1541,1542,1543],{},"你可以在 Telegram 上给它发消息让它做事，结果推送到 Discord；或者在 CLI 里开发时让它监听 Git 提交自动跑测试。",[24,1544,1545],{},"一个 Agent 实例，多个入口。",[16,1547,1548],{"id":1548},"核心能力",[92,1550,1551],{"id":1551},"自我进化",[20,1553,1554,1555,59],{},"Hermes Agent 最独特的能力是 ",[24,1556,1557],{},"Profile 系统",[493,1559,1560,1563,1566,1573],{},[99,1561,1562],{},"每个 Profile 是一个\"人格 + 技能包\"的组合",[99,1564,1565],{},"你可以创建多个 Profile（如\"代码助手\"、\"研究助手\"、\"项目经理\"）",[99,1567,1568,1569,1572],{},"Agent 在执行任务后会",[24,1570,1571],{},"自动总结经验","，更新 Profile 的技能库",[99,1574,1575],{},"下次遇到类似任务，直接调用已有技能，不用从零开始",[20,1577,1578],{},"这意味着 Hermes Agent 不是\"每次都从零开始的 ChatBot\"，而是\"会积累经验的数字员工\"。",[92,1580,1581],{"id":1581},"工具调用",[20,1583,1584],{},"Hermes Agent 支持自定义工具（function calling）：",[493,1586,1587,1590,1593,1596,1599,1602],{},[99,1588,1589],{},"搜索引擎（Google \u002F Bing \u002F SearXNG）",[99,1591,1592],{},"代码执行（Python sandbox）",[99,1594,1595],{},"文件读写",[99,1597,1598],{},"Web 浏览（Playwright）",[99,1600,1601],{},"自定义 API 调用",[99,1603,1604],{},"数据库查询",[20,1606,1607],{},"工具配置是 JSON 格式，添加新工具只需写一个 function 定义。",[92,1609,1610],{"id":1610},"记忆管理",[20,1612,1613],{},"记忆系统基于向量数据库（默认 ChromaDB）：",[493,1615,1616,1619,1622,1625],{},[99,1617,1618],{},"自动从对话中提取关键信息存入语义记忆",[99,1620,1621],{},"支持手动\"forget\"删除特定记忆",[99,1623,1624],{},"记忆有 TTL（过期时间），避免无限膨胀",[99,1626,1627],{},"支持记忆导出\u002F导入（JSON 格式），方便迁移",[16,1629,1630],{"id":1630},"使用体验",[92,1632,1633],{"id":1633},"安装部署",[801,1635,1637],{"className":803,"code":1636,"language":805,"meta":806,"style":806},"git clone https:\u002F\u002Fgithub.com\u002FNousResearch\u002Fhermes-agent\ncd hermes-agent\npip install -r requirements.txt\ncp .env.example .env\n# 配置 LLM API key、消息平台 token\npython -m hermes.agent\n",[130,1638,1639,1650,1658,1672,1684,1691],{"__ignoreMap":806},[810,1640,1641,1644,1647],{"class":812,"line":813},[810,1642,1643],{"class":816},"git",[810,1645,1646],{"class":824}," clone",[810,1648,1649],{"class":824}," https:\u002F\u002Fgithub.com\u002FNousResearch\u002Fhermes-agent\n",[810,1651,1652,1655],{"class":812,"line":1266},[810,1653,1654],{"class":820},"cd",[810,1656,1657],{"class":824}," hermes-agent\n",[810,1659,1660,1663,1666,1669],{"class":812,"line":1263},[810,1661,1662],{"class":816},"pip",[810,1664,1665],{"class":824}," install",[810,1667,1668],{"class":820}," -r",[810,1670,1671],{"class":824}," requirements.txt\n",[810,1673,1675,1678,1681],{"class":812,"line":1674},4,[810,1676,1677],{"class":816},"cp",[810,1679,1680],{"class":824}," .env.example",[810,1682,1683],{"class":824}," .env\n",[810,1685,1687],{"class":812,"line":1686},5,[810,1688,1690],{"class":1689},"sJ8bj","# 配置 LLM API key、消息平台 token\n",[810,1692,1694,1697,1700],{"class":812,"line":1693},6,[810,1695,1696],{"class":816},"python",[810,1698,1699],{"class":820}," -m",[810,1701,1702],{"class":824}," hermes.agent\n",[20,1704,1705,1706,1709,1710,1713,1714,1717],{},"部署需要 ",[24,1707,1708],{},"Python 3.11+","、",[24,1711,1712],{},"至少 8GB RAM","（跑本地模型需要更多）、",[24,1715,1716],{},"向量数据库","（默认 ChromaDB，可选 Qdrant）。",[92,1719,1720],{"id":1720},"日常使用",[20,1722,1723,1724,59],{},"最顺的使用方式是 ",[24,1725,1726],{},"Telegram + Claude API",[96,1728,1729,1732,1735,1738],{},[99,1730,1731],{},"在 Telegram 上给 Agent 发消息",[99,1733,1734],{},"Agent 读取记忆、规划任务、调用工具",[99,1736,1737],{},"执行过程中实时推送进度",[99,1739,1740],{},"完成后推送结果 + 自动更新记忆",[20,1742,1743],{},"体感类似\"有一个 7×24 小时在线的助手\"，但它不是即问即答——复杂任务可能需要 2-5 分钟。",[92,1745,1746],{"id":1746},"短板",[493,1748,1749,1755,1761,1767],{},[99,1750,1751,1754],{},[24,1752,1753],{},"文档偏英文","：几乎没有中文文档，国内用户上手门槛高",[99,1756,1757,1760],{},[24,1758,1759],{},"稳定性","：项目迭代极快（每周多个 commit），偶尔有 breaking change",[99,1762,1763,1766],{},[24,1764,1765],{},"资源消耗","：记忆系统 + 多 Agent 会占用较多内存和 API token",[99,1768,1769,1772],{},[24,1770,1771],{},"UI 简陋","：Web UI 只是基础版，不如 Manus \u002F OpenClaw 精致",[16,1774,1775],{"id":1775},"价格",[144,1777,1778,1787],{},[147,1779,1780],{},[150,1781,1782,1785],{},[153,1783,1784],{},"项目",[153,1786,628],{},[166,1788,1789,1797,1805,1813,1821],{},[150,1790,1791,1794],{},[171,1792,1793],{},"Hermes Agent 本体",[171,1795,1796],{},"免费（开源）",[150,1798,1799,1802],{},[171,1800,1801],{},"LLM API",[171,1803,1804],{},"BYOK，用 Claude\u002FGPT 按各自 API 计费",[150,1806,1807,1810],{},[171,1808,1809],{},"本地模型",[171,1811,1812],{},"免费但需要 GPU（70B 模型需 ~48GB VRAM）",[150,1814,1815,1818],{},[171,1816,1817],{},"服务器",[171,1819,1820],{},"自托管需要一台 VPS（推荐 4 核 16GB 起步）",[150,1822,1823,1826],{},[171,1824,1825],{},"消息平台",[171,1827,1828],{},"Telegram\u002FDiscord 等均免费",[20,1830,1831,1834],{},[24,1832,1833],{},"最低成本","：一台 $5\u002F月 VPS + Claude API 按量付费 ≈ $10-20\u002F月可跑日常任务。",[16,1836,1837],{"id":1837},"与同类对比",[144,1839,1840,1853],{},[147,1841,1842],{},[150,1843,1844,1846,1848,1851],{},[153,1845,155],{},[153,1847,584],{},[153,1849,1850],{},"Manus",[153,1852,581],{},[166,1854,1855,1865,1877,1889,1903,1916,1928],{},[150,1856,1857,1859,1861,1863],{},[171,1858,596],{},[171,1860,722],{},[171,1862,715],{},[171,1864,722],{},[150,1866,1867,1869,1872,1874],{},[171,1868,1551],{},[171,1870,1871],{},"✅ Profile 系统",[171,1873,715],{},[171,1875,1876],{},"⚠️ 有限",[150,1878,1879,1881,1884,1887],{},[171,1880,644],{},[171,1882,1883],{},"✅ 分层记忆",[171,1885,1886],{},"❌ 单会话",[171,1888,722],{},[150,1890,1891,1894,1897,1900],{},[171,1892,1893],{},"跨平台部署",[171,1895,1896],{},"✅ 14 渠道",[171,1898,1899],{},"❌ Web only",[171,1901,1902],{},"⚠️ 桌面+CLI",[150,1904,1905,1907,1910,1913],{},[171,1906,612],{},[171,1908,1909],{},"★★★★☆",[171,1911,1912],{},"★☆☆☆☆",[171,1914,1915],{},"★★★☆☆",[150,1917,1918,1921,1924,1926],{},[171,1919,1920],{},"中文体验",[171,1922,1923],{},"★★☆☆☆",[171,1925,1909],{},[171,1927,1915],{},[150,1929,1930,1933,1936,1939],{},[171,1931,1932],{},"适合人群",[171,1934,1935],{},"极客\u002F研究者",[171,1937,1938],{},"普通用户",[171,1940,1941],{},"开发者",[16,1943,1107],{"id":1106},[20,1945,1946,1949],{},[24,1947,1948],{},"Q：Hermes Agent 能用中文交互吗？","\n能，但体验一般。Hermes 4 模型的中文能力不如 Claude\u002FGPT，且文档和社区以英文为主。建议用 Claude\u002FGPT 作为后端模型，中文交互质量会好很多。",[20,1951,1952,1955],{},[24,1953,1954],{},"Q：和 OpenManus 有什么区别？","\nOpenManus 是 Manus 的开源复刻版，定位是\"通用 Agent 执行器\"。Hermes Agent 更侧重\"长期陪伴+自我进化+多平台部署\"。OpenManus 更轻量，Hermes 功能更全但更重。",[20,1957,1958,1961],{},[24,1959,1960],{},"Q：需要什么硬件？","\n纯 API 模式（用 Claude\u002FGPT）只需一台普通 VPS。跑本地 Hermes 4 70B 需要 ~48GB VRAM，405B 需要 ~240GB VRAM（多卡服务器）。",[20,1963,1964,1967],{},[24,1965,1966],{},"Q：记忆数据存在哪？","\n默认存在本地 ChromaDB（SQLite + 向量索引）。可以配置为 Qdrant、Weaviate 等远程向量数据库。数据完全自主，不会上传到任何第三方。",[20,1969,1970,1973],{},[24,1971,1972],{},"Q：能同时跑多个 Profile 吗？","\n可以。每个 Profile 是独立的记忆+技能库，可以并行运行。比如同时让\"代码助手\"Profile 审查代码、\"研究助手\"Profile 做市场调研。",[16,1975,1185],{"id":1185},[493,1977,1978,1984,1990,1996,2001],{},[99,1979,1980],{},[853,1981,1983],{"href":1982},"\u002Fagent\u002Fgeneral\u002Fmanus.html","Manus 工具卡",[99,1985,1986],{},[853,1987,1989],{"href":1988},"\u002Fagent\u002Fgeneral\u002Fopenmanus.html","OpenManus 工具卡",[99,1991,1992],{},[853,1993,1995],{"href":1994},"\u002Fagent\u002Fdesktop\u002Fopenclaw.html","OpenClaw 工具卡",[99,1997,1998],{},[853,1999,2000],{"href":1210},"Manus 深度评测",[99,2002,2003],{},[853,2004,2006],{"href":2005},"\u002Fcompare\u002Fmanus-vs-genspark.html","Manus vs Genspark 对比",[1259,2008,2009],{},"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 .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}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":806,"searchDepth":1263,"depth":1263,"links":2011},[2012,2013,2018,2023,2028,2029,2030,2031],{"id":18,"depth":1266,"text":18},{"id":1377,"depth":1266,"text":1378,"children":2014},[2015,2016,2017],{"id":1381,"depth":1263,"text":1382},{"id":1424,"depth":1263,"text":1425},{"id":1452,"depth":1263,"text":1453},{"id":1548,"depth":1266,"text":1548,"children":2019},[2020,2021,2022],{"id":1551,"depth":1263,"text":1551},{"id":1581,"depth":1263,"text":1581},{"id":1610,"depth":1263,"text":1610},{"id":1630,"depth":1266,"text":1630,"children":2024},[2025,2026,2027],{"id":1633,"depth":1263,"text":1633},{"id":1720,"depth":1263,"text":1720},{"id":1746,"depth":1263,"text":1746},{"id":1775,"depth":1266,"text":1775},{"id":1837,"depth":1266,"text":1837},{"id":1106,"depth":1266,"text":1107},{"id":1185,"depth":1266,"text":1185},"general","\u002Fimg\u002Ftools\u002Fhermes-agent.webp","Hermes Agent 真实评测：Nous Research 出品的开源自进化 AI Agent，2026 年 2 月开源即获 10 万 GitHub Star。MoA 混合智能体架构、记忆系统、技能自动创建、跨平台部署（Telegram\u002FDiscord\u002FSlack\u002FWhatsApp\u002FCLI）。本文整理核心能力、使用体验、与 Manus\u002FOpenClaw 对比、适用场景。",false,"https:\u002F\u002Fgithub.com\u002FNousResearch\u002Fhermes-agent",[2038],"en",{},[2041,2042,2043,2044],"hermes-4-405b","hermes-4-70b","claude-sonnet-4","gpt-5",[2046,2047,2048,2049],"需要中文为主交互的玩家（文档和界面均为英文）","不想折腾部署和配置的个人用户","需要生产级稳定性（项目仍在快速迭代）","预算有限且没有 GPU 服务器的用户","\u002Ftools\u002Fagent\u002Fgeneral\u002Fhermes-agent","agent",[2053,2054,2055],"linux","macos","windows",[2057,2062],{"plan":2058,"price":989,"limit":2059,"cn_pay":2060,"note":2061},"开源版","完整功能，自托管","—","BYOK 模式",{"plan":2063,"price":2064,"limit":2065,"cn_pay":2066,"note":2067},"API（Hermes 4）","按 token 计费","405B 模型按量付费","⚠️ 需海外卡","不想自托管时","免费（开源，BYOK）","2026-07-04",[2071],"openhuman-deep-review",{"power":1674,"ux":1263,"price":1686,"cn_support":1266,"stability":1263},{"title":584,"description":2034},"Hermes Agent 评测 2026：Nous Research 开源自进化 AI Agent，10 万 Star",[2076,2078,2081,2084,2087],{"title":2077,"url":2036},"Hermes Agent GitHub",{"title":2079,"url":2080},"Nous Research 官网","https:\u002F\u002Fnousresearch.com",{"title":2082,"url":2083},"Hermes Agent 安装教程","https:\u002F\u002Fblog.csdn.net\u002Fyweng18\u002Farticle\u002Fdetails\u002F161148047",{"title":2085,"url":2086},"Hermes MoA 体验报告","https:\u002F\u002Fm.toutiao.com\u002Fgroup\u002F7657411078791365171\u002F",{"title":2088,"url":2089},"Hermes vs OpenCode 对比","https:\u002F\u002Fm.toutiao.com\u002Fgroup\u002F7645892543409816064\u002F","tools\u002Fagent\u002Fgeneral\u002Fhermes-agent",[2092,2093,2094,2095,2096],"需要自托管、数据完全自主的 AI Agent","对 Agent 自我进化 \u002F 记忆系统有研究兴趣","需要跨平台部署（Telegram \u002F Discord \u002F Slack 等多通道）","有 GPU 服务器可以跑 Hermes 4 模型","想要一个长期陪伴型个人 Agent","Nous Research 开源自进化 AI Agent，10 万 Star，MoA 混合智能体",[2099,2100,2101,2102,2103,2104,2105],"general-agent","autonomous","opensource","self-evolving","memory","nous-research","moa","2026 最火开源 Agent。MoA 混合智能体+自我进化+记忆系统+跨平台部署，GitHub 10 万 Star。适合长期陪伴型任务和自托管 Agent 研究，但中文支持和文档偏弱。","3gxu2kQEY9RgT_Ir0N2dPCaoLfjN4jFfYElq7IKZGhA",{"id":2109,"title":589,"alternatives":2110,"api_compatible":1321,"body":2112,"category":2032,"chinese_friendly":1263,"cover":2633,"description":2634,"domestic":2035,"extension":1320,"faq":2635,"free":2035,"github":1321,"languages":2648,"lastVerified":1321,"meta":2650,"models":1321,"navigation":1323,"notSuitable":1321,"opensource":1323,"path":2651,"pillar":2051,"platforms":2652,"priceTable":2653,"pricing":2661,"published":1325,"relatedPlaybooks":2662,"relatedReviews":2664,"score":2665,"self_host":1323,"seo":2666,"seoTitle":1321,"slug":1327,"sources":2667,"stem":2674,"suitable":1321,"tagline":2675,"tags":2676,"updated":1325,"verdict":2682,"website":855,"__hash__":2683},"tools\u002Ftools\u002Fagent\u002Fgeneral\u002Fopenhuman.md",[1329,1328,2111],"agent\u002Fplatform\u002Fcoze",{"type":13,"value":2113,"toc":2621},[2114,2118,2121,2124,2126,2206,2208,2224,2228,2233,2256,2261,2284,2287,2332,2338,2341,2352,2355,2494,2497,2540,2544,2573,2575,2594,2596,2618],[16,2115,2117],{"id":2116},"tldr","TL;DR",[20,2119,2120],{},"OpenHuman 是 TinyHumans 团队 2026-05 推出的开源个人 AI 超级智能助手，7.8k+ GitHub stars。差异点：Rust（Tauri）桌面优先架构 + 118+ 第三方 OAuth 集成 + Memory Tree 长期记忆系统（10 亿 token 容量）+ TokenJuice 智能压缩（省 80% token）+ 智能模型路由 + 桌面吉祥物 + Google Meet 参会 + 语音交互 + Obsidian 知识库兼容。20 分钟自动同步你的数字生活，构建本地私有的个人记忆库。",[20,2122,2123],{},"适合：追求个人 AI 助手真正了解你的知识工作者；Obsidian 用户；跨越多个工具的协作场景；关注数据隐私的用户。不适合：追求极致轻量单功能工具；需要完全离线运行（OAuth 和模型调用需网络）；无法接受 Early Beta 产品的不稳定。",[16,2125,1548],{"id":1548},[493,2127,2128,2134,2140,2146,2152,2157,2162,2167,2172,2177,2183,2188,2194,2200],{},[99,2129,2130,2133],{},[24,2131,2132],{},"Memory Tree 长期记忆系统","：自动抓取邮件、文档、聊天记录 → 智能评分 → 层级摘要树 → 本地 SQLite 存储 + Obsidian 兼容 .md 文件",[99,2135,2136,2139],{},[24,2137,2138],{},"118+ 第三方 OAuth 集成","：Gmail \u002F Outlook \u002F Notion \u002F GitHub \u002F Slack \u002F Google Calendar \u002F Stripe \u002F Linear \u002F Jira 等",[99,2141,2142,2145],{},[24,2143,2144],{},"TokenJuice 智能压缩","：HTML→Markdown \u002F URL 缩短 \u002F 去重，最高省 80% token",[99,2147,2148,2151],{},[24,2149,2150],{},"智能模型路由","：统一订阅，自动分配推理 \u002F 快速 \u002F 多模态 \u002F 本地模型",[99,2153,2154,2156],{},[24,2155,712],{},"：有表情、会说话的桌面小伙伴，响应环境变化",[99,2158,2159,2161],{},[24,2160,541],{},"：以真实参与者身份加入线上会议",[99,2163,2164,2166],{},[24,2165,547],{},"：STT 语音输入 + ElevenLabs TTS 语音输出 + 唇形同步",[99,2168,2169,2171],{},[24,2170,553],{},"：即使不主动交互，也在后台分析和整理信息",[99,2173,2174,2176],{},[24,2175,681],{},"：每 20 分钟自动遍历所有活跃连接，拉取最新数据",[99,2178,2179,2182],{},[24,2180,2181],{},"Obsidian 兼容","：记忆以 Markdown 形式存储，可直接用 Obsidian 查看\u002F编辑",[99,2184,2185,2187],{},[24,2186,247],{},"：核心数据存储在本地 SQLite，不上传云端",[99,2189,2190,2193],{},[24,2191,2192],{},"本地加密","：数据在设备端加密存储",[99,2195,2196,2199],{},[24,2197,2198],{},"Ollama 支持","：可运行本地 LLM，敏感任务完全不上云",[99,2201,2202,2205],{},[24,2203,2204],{},"GPL-3.0 开源","：完整源码可审计、可定制",[16,2207,1775],{"id":1775},[493,2209,2210,2215,2221],{},[99,2211,2212,2214],{},[24,2213,975],{},"：价格待定（Early Beta）；一价全包所有模型 + 自动路由",[99,2216,2217,2220],{},[24,2218,2219],{},"自托管","：$0；GPL-3.0 协议，需自备 LLM API 或本地 Ollama 模型",[99,2222,2223],{},"自托管一次中等任务 API 费用 $0.02-0.5（取决于模型选择）",[16,2225,2227],{"id":2226},"实测个人-ai-助手场景","实测（个人 AI 助手场景）",[20,2229,2230],{},[24,2231,2232],{},"亮点：",[493,2234,2235,2238,2241,2244,2247,2250,2253],{},[99,2236,2237],{},"Memory Tree 让 AI 真正拥有\"长期记忆\"，不是每次对话从零开始",[99,2239,2240],{},"118+ 集成开箱即用，OAuth 一键连接，无需手动配置 API",[99,2242,2243],{},"TokenJuice 压缩效果显著，实测 token 消耗降低 60-80%",[99,2245,2246],{},"桌面 UI 设计精美，吉祥物交互有趣，不是命令行工具",[99,2248,2249],{},"自动同步机制省心，不用手动通知 AI 新信息",[99,2251,2252],{},"Obsidian 兼容让记忆透明可读，用户完全掌控自己的数据",[99,2254,2255],{},"智能模型路由省去手动切换模型的麻烦",[20,2257,2258],{},[24,2259,2260],{},"踩坑：",[493,2262,2263,2266,2269,2272,2275,2278,2281],{},[99,2264,2265],{},"Early Beta 阶段，功能迭代快，偶尔有 breaking change",[99,2267,2268],{},"中文支持依赖底层模型，部分场景效果不如英文",[99,2270,2271],{},"OAuth 连接器稳定性参差不齐，部分服务偶尔断连",[99,2273,2274],{},"桌面吉祥物 CPU\u002F内存占用不低，低配机器略卡",[99,2276,2277],{},"文档尚在完善中，部分功能缺少详细说明",[99,2279,2280],{},"自动同步频率固定 20 分钟，无法手动触发即时同步",[99,2282,2283],{},"隐私边界需关注：虽然数据本地存储，但 OAuth 连接和模型调用涉及网络",[16,2285,2286],{"id":2286},"上手",[801,2288,2290],{"className":803,"code":2289,"language":805,"meta":806,"style":806},"# macOS \u002F Linux\ncurl -fsSL https:\u002F\u002Fraw.githubusercontent.com\u002Ftinyhumansai\u002Fopenhuman\u002Fmain\u002Fscripts\u002Finstall.sh | bash\n\n# Windows (PowerShell)\nirm https:\u002F\u002Fraw.githubusercontent.com\u002Ftinyhumansai\u002Fopenhuman\u002Fmain\u002Fscripts\u002Finstall.ps1 | iex\n",[130,2291,2292,2297,2309,2314,2319],{"__ignoreMap":806},[810,2293,2294],{"class":812,"line":813},[810,2295,2296],{"class":1689},"# macOS \u002F Linux\n",[810,2298,2299,2301,2303,2305,2307],{"class":812,"line":1266},[810,2300,817],{"class":816},[810,2302,821],{"class":820},[810,2304,825],{"class":824},[810,2306,829],{"class":828},[810,2308,832],{"class":816},[810,2310,2311],{"class":812,"line":1263},[810,2312,2313],{"emptyLinePlaceholder":1323},"\n",[810,2315,2316],{"class":812,"line":1674},[810,2317,2318],{"class":1689},"# Windows (PowerShell)\n",[810,2320,2321,2324,2327,2329],{"class":812,"line":1686},[810,2322,2323],{"class":816},"irm",[810,2325,2326],{"class":824}," https:\u002F\u002Fraw.githubusercontent.com\u002Ftinyhumansai\u002Fopenhuman\u002Fmain\u002Fscripts\u002Finstall.ps1",[810,2328,829],{"class":828},[810,2330,2331],{"class":816}," iex\n",[20,2333,2334,2335],{},"或从官网下载安装包：",[853,2336,855],{"href":855,"rel":2337},[857],[20,2339,2340],{},"首次启动后：",[96,2342,2343,2346,2349],{},[99,2344,2345],{},"完成 OAuth 授权连接你的邮箱 \u002F 日历 \u002F 文档 \u002F 代码仓库",[99,2347,2348],{},"等待 2-5 分钟初始数据同步（Memory Tree 自动构建）",[99,2350,2351],{},"开始对话——AI 已经知道你的工作背景和习惯",[16,2353,2354],{"id":2354},"对比",[144,2356,2357,2371],{},[147,2358,2359],{},[150,2360,2361,2363,2365,2367,2369],{},[153,2362,155],{},[153,2364,589],{},[153,2366,578],{},[153,2368,581],{},[153,2370,584],{},[166,2372,2373,2389,2401,2416,2429,2444,2457,2470,2482],{},[150,2374,2375,2378,2381,2384,2387],{},[171,2376,2377],{},"形态",[171,2379,2380],{},"桌面应用",[171,2382,2383],{},"桌面+CLI",[171,2385,2386],{},"终端",[171,2388,2386],{},[150,2390,2391,2393,2395,2397,2399],{},[171,2392,596],{},[171,2394,607],{},[171,2396,599],{},[171,2398,602],{},[171,2400,602],{},[150,2402,2403,2405,2408,2411,2414],{},[171,2404,612],{},[171,2406,2407],{},"✅ UI 优先，几分钟",[171,2409,2410],{},"✅ 桌面",[171,2412,2413],{},"⚠️ 终端",[171,2415,2413],{},[150,2417,2418,2420,2423,2425,2427],{},[171,2419,644],{},[171,2421,2422],{},"✅ Memory Tree",[171,2424,647],{},[171,2426,650],{},[171,2428,653],{},[150,2430,2431,2433,2436,2439,2442],{},[171,2432,663],{},[171,2434,2435],{},"118+ OAuth",[171,2437,2438],{},"少",[171,2440,2441],{},"需自建",[171,2443,2441],{},[150,2445,2446,2448,2451,2453,2455],{},[171,2447,681],{},[171,2449,2450],{},"✅ 20分钟",[171,2452,715],{},[171,2454,715],{},[171,2456,715],{},[150,2458,2459,2461,2464,2466,2468],{},[171,2460,696],{},[171,2462,2463],{},"✅ 内置智能",[171,2465,699],{},[171,2467,702],{},[171,2469,702],{},[150,2471,2472,2474,2476,2478,2480],{},[171,2473,712],{},[171,2475,722],{},[171,2477,715],{},[171,2479,715],{},[171,2481,715],{},[150,2483,2484,2486,2488,2490,2492],{},[171,2485,236],{},[171,2487,247],{},[171,2489,239],{},[171,2491,731],{},[171,2493,731],{},[16,2495,2496],{"id":2496},"避坑",[493,2498,2499,2505,2511,2517,2523,2528,2534],{},[99,2500,2501,2504],{},[24,2502,2503],{},"Early Beta 不稳定","：功能迭代快，关键工作建议备份记忆数据",[99,2506,2507,2510],{},[24,2508,2509],{},"中文效果","：依赖底层模型，GPT-4o \u002F Claude 中文效果好，本地模型待验证",[99,2512,2513,2516],{},[24,2514,2515],{},"OAuth 断连","：部分服务 token 会过期，需定期检查连接状态",[99,2518,2519,2522],{},[24,2520,2521],{},"TokenJuice 精度","：压缩虽好，处理合同\u002F账单等敏感内容需注意细节丢失",[99,2524,2525,2527],{},[24,2526,557],{},"：桌面吉祥物 + 后台同步建议 8GB+ 内存",[99,2529,2530,2533],{},[24,2531,2532],{},"网络依赖","：OAuth 连接和远程模型调用需要稳定网络",[99,2535,2536,2539],{},[24,2537,2538],{},"隐私边界","：虽然数据本地存储，但 OAuth 连接意味着服务商可知你的连接状态",[16,2541,2543],{"id":2542},"适合-不适合","适合 \u002F 不适合",[493,2545,2546,2549,2552,2555,2558,2561,2564,2567,2570],{},[99,2547,2548],{},"✅ 追求个人 AI 助手真正了解你的知识工作者",[99,2550,2551],{},"✅ Obsidian 用户，希望 AI 自动维护知识库",[99,2553,2554],{},"✅ 工作流跨越多个工具（Gmail \u002F Slack \u002F Notion \u002F GitHub）",[99,2556,2557],{},"✅ 关注数据隐私，希望敏感信息保留在本地",[99,2559,2560],{},"✅ 愿意尝试 Early Beta，能接受快速迭代",[99,2562,2563],{},"❌ 追求极致轻量，不需要那么多集成功能",[99,2565,2566],{},"❌ 对隐私要求极高，希望完全离线运行",[99,2568,2569],{},"❌ 无法接受 Beta 产品的粗糙边缘和偶发 bug",[99,2571,2572],{},"❌ 需要中文深度优化（当前英文体验最佳）",[16,2574,1185],{"id":1185},[493,2576,2577,2582,2588],{},[99,2578,2579],{},[853,2580,2581],{"href":1988},"OpenManus 工具卡：开源版 Manus",[99,2583,2584],{},[853,2585,2587],{"href":2586},"\u002Fagent\u002Fgeneral\u002Fhermes-agent.html","Hermes Agent 工具卡：自学习 AI Agent",[99,2589,2590],{},[853,2591,2593],{"href":2592},"\u002Fagent\u002Fplatform\u002Fcoze.html","Coze 平台评测",[16,2595,1214],{"id":1214},[96,2597,2598,2603,2608,2613],{},[99,2599,1219,2600],{},[853,2601,1222],{"href":1222,"rel":2602},[857],[99,2604,1226,2605],{},[853,2606,855],{"href":855,"rel":2607},[857],[99,2609,1232,2610],{},[853,2611,1235],{"href":1235,"rel":2612},[857],[99,2614,1239,2615],{},[853,2616,1242],{"href":1242,"rel":2617},[857],[1259,2619,2620],{},"html pre.shiki code .sJ8bj, html code.shiki .sJ8bj{--shiki-default:#6A737D;--shiki-dark:#6A737D}html pre.shiki code .sScJk, html code.shiki .sScJk{--shiki-default:#6F42C1;--shiki-dark:#B392F0}html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .szBVR, html code.shiki .szBVR{--shiki-default:#D73A49;--shiki-dark:#F97583}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":806,"searchDepth":1263,"depth":1263,"links":2622},[2623,2624,2625,2626,2627,2628,2629,2630,2631,2632],{"id":2116,"depth":1266,"text":2117},{"id":1548,"depth":1266,"text":1548},{"id":1775,"depth":1266,"text":1775},{"id":2226,"depth":1266,"text":2227},{"id":2286,"depth":1266,"text":2286},{"id":2354,"depth":1266,"text":2354},{"id":2496,"depth":1266,"text":2496},{"id":2542,"depth":1266,"text":2543},{"id":1185,"depth":1266,"text":1185},{"id":1214,"depth":1266,"text":1214},"\u002Fimg\u002Ftools\u002Fopenhuman.svg","OpenHuman 真实评测：TinyHumans 团队 2026-05 开源的个人 AI 超级智能助手，7.8k+ GitHub stars。GPL-3.0 协议 + Rust（Tauri）架构 + 桌面 UI 优先 + 118+ 第三方 OAuth 集成 + Memory Tree 长期记忆系统（10亿 token 容量）+ TokenJuice 智能压缩（省 80% token）+ 智能模型路由 + 桌面吉祥物 + Google Meet 参会 + 语音交互 + Obsidian 知识库兼容。20 分钟自动同步你的邮箱、日历、文档、代码仓库，构建本地私有的个人记忆库。",[2636,2638,2639,2642,2645],{"q":1111,"a":2637},"ChatGPT \u002F Claude 是通用对话 AI，每次对话从零开始，没有你的长期记忆。OpenHuman 通过 Memory Tree 系统持续学习你的邮件、日历、文档、代码仓库，构建一个持续更新的个人记忆库。它知道『你是谁』而非只是『你问了什么』。",{"q":1118,"a":1121},{"q":2640,"a":2641},"Memory Tree 是什么？","OpenHuman 的核心记忆系统。所有接入数据（邮件、文档、聊天记录）被转化为 ≤3000 token 的 Markdown 块，经智能评分后折叠成层级摘要树，存入本地 SQLite。同时生成 .md 文件同步到 Obsidian 知识库。记忆容量可达 10 亿 token。",{"q":2643,"a":2644},"TokenJuice 有什么用？","TokenJuice 是 OpenHuman 的智能压缩层，在数据进入 LLM 前进行预处理——HTML 转 Markdown、长 URL 缩短、重复内容去重等，最高可降低 80% 的 token 消耗，大幅降低 API 成本和响应延迟。",{"q":2646,"a":2647},"支持哪些第三方服务？","118+ 第三方 OAuth 集成，涵盖：Gmail \u002F Outlook（邮件）、Notion \u002F Google Docs \u002F Obsidian（文档）、GitHub \u002F GitLab（代码）、Slack \u002F Discord（沟通）、Google Calendar \u002F Outlook Calendar（日历）、Google Drive \u002F Dropbox（存储）、Stripe（支付）、Linear \u002F Jira（项目）等。",[2038,2649],"multi",{},"\u002Ftools\u002Fagent\u002Fgeneral\u002Fopenhuman",[2053,2054,2055],[2654,2658],{"plan":975,"price":2655,"features":2656,"notes":2657},"待定","全部模型 + 自动路由 + 118+ 集成 + Memory Tree + TokenJuice","Early Beta 定价可能调整",{"plan":2219,"price":989,"features":2659,"notes":2660},"GPL-3.0 协议 + 全部功能 + 可接 Ollama 本地模型","需自备 LLM API","统一订阅制（一价全包，自动模型路由）\u002F 自备 LLM API 可选",[2663],"onboarding\u002Fpersonal-ai-agent",[2071],{"power":1674,"ux":1686,"price":1674,"cn_support":1263,"stability":1263},{"title":589,"description":2634},[2668,2670,2672],{"name":2669,"url":1222,"accessed":1325},"OpenHuman GitHub（TinyHumans 组织）",{"name":2671,"url":855,"accessed":1325},"OpenHuman 官网",{"name":2673,"url":1235,"accessed":1325},"OpenHuman 文档","tools\u002Fagent\u002Fgeneral\u002Fopenhuman","TinyHumans 团队开源个人 AI 超级智能——118+ 集成 \u002F Memory Tree \u002F TokenJuice 压缩 \u002F 桌面优先",[2101,2677,2678,2051,2679,2680,2681],"personal-ai","memory-tree","desktop","obsidian","rust","想要一个真正懂你、记得你、能自动学习的个人 AI 助手——OpenHuman 是目前开源社区最完整的选择。Memory Tree + 118+ 集成 + TokenJuice 压缩 + 桌面 UI 设计，让 AI 从『聊天工具』升级为『数字分身』。适合注重隐私、使用多平台工具的知识工作者。Early Beta 阶段，功能迭代快，适合愿意尝鲜的用户。","04_vvWTbB6SIvLrj3ZPMYuag3r930_UxuzpUgV7DQjI",{"id":2685,"title":2686,"alternatives":2687,"api_compatible":1321,"body":2691,"category":2032,"chinese_friendly":1674,"cover":3343,"description":3344,"domestic":2035,"extension":1320,"faq":3345,"free":2035,"github":1321,"languages":3358,"lastVerified":1321,"meta":3360,"models":1321,"navigation":1323,"notSuitable":1321,"opensource":1323,"path":3361,"pillar":2051,"platforms":3362,"priceTable":3364,"pricing":3368,"published":3369,"relatedPlaybooks":3370,"relatedReviews":3372,"score":3374,"self_host":1323,"seo":3375,"seoTitle":1321,"slug":1329,"sources":3376,"stem":3386,"suitable":1321,"tagline":3387,"tags":3388,"updated":3379,"verdict":3394,"website":3305,"__hash__":3395},"tools\u002Ftools\u002Fagent\u002Fgeneral\u002Fopenmanus.md","OpenManus",[2688,2689,2690],"agent\u002Fplatform\u002Flangflow","agent\u002Fplatform\u002Fn8n","agent\u002Fprotocol\u002Fcomposio",{"type":13,"value":2692,"toc":3331},[2693,2695,2702,2705,2707,2783,2785,2802,2806,2810,2836,2840,2866,2868,3000,3003,3011,3013,3184,3186,3246,3248,3274,3276,3296,3298,3328],[16,2694,2117],{"id":2116},[20,2696,2697,2698,2701],{},"OpenManus 是 MetaGPT 核心团队 2025-03 推出的开源 Manus 替代方案，52,000+ GitHub stars。差异点：MIT 协议 + Python 模块化架构 + 多 agent orchestration（",[130,2699,2700],{},"run_flow.py","）+ Playwright 浏览器自动化 + MCP 工具协议支持 + DataAnalysis 内置模式 + OpenManus-RL 强化学习分支 + 自定义工具基类（BaseTool）+ 多模型（GPT-4o \u002F Claude \u002F Qwen VL Plus）。零邀请码、零订阅、零供应商绑定。",[20,2703,2704],{},"适合：想自托管复刻 Manus 体验的开发者；研究 \u002F 学术 \u002F 教育用通用 agent 实现学习；隐私敏感 + 不愿数据上 Manus 商业云；预算紧（只付 LLM API）；中国大陆开发者（搭配 Qwen \u002F DeepSeek 本地化）。不适合：非开发者 \u002F 不会折腾 Python + Playwright；要 GUI \u002F 上手即用；生产级稳定（项目演进快，文档滞后）。",[16,2706,1548],{"id":1548},[493,2708,2709,2717,2723,2729,2735,2741,2747,2753,2759,2765,2771,2777],{},[99,2710,2711,59,2714,2716],{},[24,2712,2713],{},"多 agent orchestration",[130,2715,2700],{}," 编排多个专门 agent 协作",[99,2718,2719,2722],{},[24,2720,2721],{},"Playwright 浏览器自动化","：截图 + DOM 操作 + 表单填写 + 信息抓取",[99,2724,2725,2728],{},[24,2726,2727],{},"MCP 协议支持","：可调用 MCP server（filesystem \u002F GitHub \u002F Postgres）",[99,2730,2731,2734],{},[24,2732,2733],{},"DataAnalysis 模式","：内置 CSV \u002F 数据分析 agent",[99,2736,2737,2740],{},[24,2738,2739],{},"OpenManus-RL","：强化学习微调分支",[99,2742,2743,2746],{},[24,2744,2745],{},"BaseTool 自定义工具","：Python 继承基类快速添加新工具",[99,2748,2749,2752],{},[24,2750,2751],{},"多模态","：文本 + 视觉输入 + 浏览器截图回环",[99,2754,2755,2758],{},[24,2756,2757],{},"多 LLM provider","：GPT-4o \u002F Claude 3.5 \u002F Qwen VL Plus \u002F DeepSeek \u002F Gemini",[99,2760,2761,2764],{},[24,2762,2763],{},"核心 agent 引擎","：reasoning + planning + execution 三阶段",[99,2766,2767,2770],{},[24,2768,2769],{},"Web UI 监控","：实时看 AI thinking process",[99,2772,2773,2776],{},[24,2774,2775],{},"任务可视化","：步骤拆解 + 执行树",[99,2778,2779,2782],{},[24,2780,2781],{},"MIT 协议","：个人 + 商用全免费",[16,2784,1775],{"id":1775},[493,2786,2787,2793,2796,2799],{},[99,2788,2789,2792],{},[24,2790,2791],{},"Free \u002F OSS","：$0；MIT 协议",[99,2794,2795],{},"真实成本 = LLM API（GPT-4o ~$5\u002FM input + $15\u002FM output \u002F Claude \u002F Qwen 等）",[99,2797,2798],{},"一次中等任务（10-20 步）API 费用 $0.05-0.5",[99,2800,2801],{},"本地 Qwen2.5 32B \u002F DeepSeek 走 vLLM \u002F Ollama 路径 $0",[16,2803,2805],{"id":2804},"实测开发者-自托管-研究场景","实测（开发者 \u002F 自托管 \u002F 研究场景）",[20,2807,2808],{},[24,2809,2232],{},[493,2811,2812,2815,2818,2821,2824,2827,2830,2833],{},[99,2813,2814],{},"52k stars 印证社区认同度 + 活跃度",[99,2816,2817],{},"MetaGPT 团队背景保证架构质量",[99,2819,2820],{},"多 agent 协作场景比单 agent 实现稳得多",[99,2822,2823],{},"Playwright 浏览器自动化非常完整",[99,2825,2826],{},"MCP 协议接入打通 Claude \u002F Cursor 生态",[99,2828,2829],{},"DataAnalysis 内置 agent 模式开箱即用",[99,2831,2832],{},"中文支持自然（Qwen VL Plus 接入）",[99,2834,2835],{},"自托管 + 数据本地，隐私 \u002F 合规友好",[20,2837,2838],{},[24,2839,2260],{},[493,2841,2842,2845,2848,2851,2854,2857,2860,2863],{},[99,2843,2844],{},"项目演进快，breaking change 偶发（pin commit 跑生产）",[99,2846,2847],{},"文档滞后新功能 1-2 个月",[99,2849,2850],{},"无官方 GUI，监控 UI 在做但不完整",[99,2852,2853],{},"需要 Python 3.12+ + Playwright 依赖（首次安装 chromium 慢）",[99,2855,2856],{},"LLM API 配置非平凡（多 provider \u002F key \u002F 模型选择）",[99,2858,2859],{},"浏览器任务遇到 CAPTCHA \u002F 反爬偶尔卡死",[99,2861,2862],{},"中文 prompt 效果依赖底层模型",[99,2864,2865],{},"生产部署需自己加监控 \u002F 错误恢复 \u002F 重试",[16,2867,2286],{"id":2286},[801,2869,2871],{"className":803,"code":2870,"language":805,"meta":806,"style":806},"git clone https:\u002F\u002Fgithub.com\u002FFoundationAgents\u002FOpenManus.git\ncd OpenManus\npython3.12 -m venv .venv && source .venv\u002Fbin\u002Factivate  # 或 .venv\\Scripts\\activate\npip install -r requirements.txt\nplaywright install chromium\n\n# 配 LLM API\ncp config\u002Fconfig.example.toml config\u002Fconfig.toml\n# 编辑 config.toml 填 OPENAI \u002F ANTHROPIC \u002F DEEPSEEK key\n\n# 单 agent 模式\npython main.py\n\n# 多 agent 模式\npython run_flow.py\n",[130,2872,2873,2882,2889,2915,2925,2935,2939,2945,2956,2962,2967,2973,2981,2986,2992],{"__ignoreMap":806},[810,2874,2875,2877,2879],{"class":812,"line":813},[810,2876,1643],{"class":816},[810,2878,1646],{"class":824},[810,2880,2881],{"class":824}," https:\u002F\u002Fgithub.com\u002FFoundationAgents\u002FOpenManus.git\n",[810,2883,2884,2886],{"class":812,"line":1266},[810,2885,1654],{"class":820},[810,2887,2888],{"class":824}," OpenManus\n",[810,2890,2891,2894,2896,2899,2902,2906,2909,2912],{"class":812,"line":1263},[810,2892,2893],{"class":816},"python3.12",[810,2895,1699],{"class":820},[810,2897,2898],{"class":824}," venv",[810,2900,2901],{"class":824}," .venv",[810,2903,2905],{"class":2904},"sVt8B"," && ",[810,2907,2908],{"class":820},"source",[810,2910,2911],{"class":824}," .venv\u002Fbin\u002Factivate",[810,2913,2914],{"class":1689},"  # 或 .venv\\Scripts\\activate\n",[810,2916,2917,2919,2921,2923],{"class":812,"line":1674},[810,2918,1662],{"class":816},[810,2920,1665],{"class":824},[810,2922,1668],{"class":820},[810,2924,1671],{"class":824},[810,2926,2927,2930,2932],{"class":812,"line":1686},[810,2928,2929],{"class":816},"playwright",[810,2931,1665],{"class":824},[810,2933,2934],{"class":824}," chromium\n",[810,2936,2937],{"class":812,"line":1693},[810,2938,2313],{"emptyLinePlaceholder":1323},[810,2940,2942],{"class":812,"line":2941},7,[810,2943,2944],{"class":1689},"# 配 LLM API\n",[810,2946,2948,2950,2953],{"class":812,"line":2947},8,[810,2949,1677],{"class":816},[810,2951,2952],{"class":824}," config\u002Fconfig.example.toml",[810,2954,2955],{"class":824}," config\u002Fconfig.toml\n",[810,2957,2959],{"class":812,"line":2958},9,[810,2960,2961],{"class":1689},"# 编辑 config.toml 填 OPENAI \u002F ANTHROPIC \u002F DEEPSEEK key\n",[810,2963,2965],{"class":812,"line":2964},10,[810,2966,2313],{"emptyLinePlaceholder":1323},[810,2968,2970],{"class":812,"line":2969},11,[810,2971,2972],{"class":1689},"# 单 agent 模式\n",[810,2974,2976,2978],{"class":812,"line":2975},12,[810,2977,1696],{"class":816},[810,2979,2980],{"class":824}," main.py\n",[810,2982,2984],{"class":812,"line":2983},13,[810,2985,2313],{"emptyLinePlaceholder":1323},[810,2987,2989],{"class":812,"line":2988},14,[810,2990,2991],{"class":1689},"# 多 agent 模式\n",[810,2993,2995,2997],{"class":812,"line":2994},15,[810,2996,1696],{"class":816},[810,2998,2999],{"class":824}," run_flow.py\n",[20,3001,3002],{},"试任务示例：",[801,3004,3009],{"className":3005,"code":3007,"language":3008},[3006],"language-text","> 帮我做一份『2026 开源 AI Agent 框架』竞品对比，含表格 + 引用 + 趋势分析，输出为 markdown 文件\n","text",[130,3010,3007],{"__ignoreMap":806},[16,3012,2354],{"id":2354},[144,3014,3015,3032],{},[147,3016,3017],{},[150,3018,3019,3021,3023,3026,3029],{},[153,3020,155],{},[153,3022,2686],{},[153,3024,3025],{},"LangChain",[153,3027,3028],{},"AutoGPT",[153,3030,3031],{},"CrewAI",[166,3033,3034,3050,3066,3079,3095,3108,3122,3136,3153,3167],{},[150,3035,3036,3038,3041,3044,3047],{},[171,3037,2377],{},[171,3039,3040],{},"现成 agent 实现",[171,3042,3043],{},"building blocks",[171,3045,3046],{},"早期通用 agent",[171,3048,3049],{},"多 agent 框架",[150,3051,3052,3055,3058,3061,3064],{},[171,3053,3054],{},"浏览器自动化",[171,3056,3057],{},"✅ Playwright",[171,3059,3060],{},"–",[171,3062,3063],{},"部分",[171,3065,3060],{},[150,3067,3068,3071,3073,3075,3077],{},[171,3069,3070],{},"MCP",[171,3072,722],{},[171,3074,3063],{},[171,3076,3060],{},[171,3078,3060],{},[150,3080,3081,3084,3087,3090,3092],{},[171,3082,3083],{},"多 agent",[171,3085,3086],{},"✅ orchestration",[171,3088,3089],{},"需自搭",[171,3091,715],{},[171,3093,3094],{},"✅ 旗舰",[150,3096,3097,3100,3102,3104,3106],{},[171,3098,3099],{},"DataAnalysis 内置",[171,3101,722],{},[171,3103,3060],{},[171,3105,3060],{},[171,3107,3060],{},[150,3109,3110,3113,3116,3118,3120],{},[171,3111,3112],{},"RL 微调",[171,3114,3115],{},"✅ OpenManus-RL",[171,3117,3060],{},[171,3119,3060],{},[171,3121,3060],{},[150,3123,3124,3127,3130,3132,3134],{},[171,3125,3126],{},"协议",[171,3128,3129],{},"MIT",[171,3131,3129],{},[171,3133,3129],{},[171,3135,3129],{},[150,3137,3138,3141,3144,3147,3150],{},[171,3139,3140],{},"Stars",[171,3142,3143],{},"52k+",[171,3145,3146],{},"100k+",[171,3148,3149],{},"170k+",[171,3151,3152],{},"30k+",[150,3154,3155,3157,3160,3163,3165],{},[171,3156,2286],{},[171,3158,3159],{},"中",[171,3161,3162],{},"难",[171,3164,3159],{},[171,3166,3159],{},[150,3168,3169,3172,3175,3178,3181],{},[171,3170,3171],{},"适合",[171,3173,3174],{},"自托管 Manus 复刻",[171,3176,3177],{},"底层 framework",[171,3179,3180],{},"学习经典",[171,3182,3183],{},"多 agent 协作",[16,3185,2496],{"id":2496},[493,3187,3188,3194,3204,3210,3216,3222,3228,3234,3240],{},[99,3189,3190,3193],{},[24,3191,3192],{},"pin commit 用生产","：项目演进快，main 分支偶尔 break",[99,3195,3196,3199,3200,3203],{},[24,3197,3198],{},"Playwright 依赖大","：首次 ",[130,3201,3202],{},"playwright install"," 下 chromium 慢，国内走镜像",[99,3205,3206,3209],{},[24,3207,3208],{},"LLM 选择","：日常用 GPT-4o-mini \u002F DeepSeek 省钱，复杂任务切 GPT-4o \u002F Claude Opus",[99,3211,3212,3215],{},[24,3213,3214],{},"本地化中文","：Qwen2.5 VL 32B + vLLM 部署可全本地 + 零成本",[99,3217,3218,3221],{},[24,3219,3220],{},"监控自加","：生产部署要加 prometheus + 错误重试 + 任务超时",[99,3223,3224,3227],{},[24,3225,3226],{},"浏览器反爬","：CAPTCHA 场景搭配 2captcha \u002F human-in-loop",[99,3229,3230,3233],{},[24,3231,3232],{},"OpenManus-RL 分支","：研究场景才需要，普通用户主仓库就够",[99,3235,3236,3239],{},[24,3237,3238],{},"MCP server","：信任来源很重要，能访问的目录 \u002F 工具要审慎",[99,3241,3242,3245],{},[24,3243,3244],{},"多 agent runaway","：复杂任务设 max_steps 防止失控烧 token",[16,3247,2543],{"id":2542},[493,3249,3250,3253,3256,3259,3262,3265,3268,3271],{},[99,3251,3252],{},"✅ 开发者 + 想自托管 Manus 风格 agent",[99,3254,3255],{},"✅ 研究 \u002F 学术 \u002F 教育用通用 agent 学习",[99,3257,3258],{},"✅ 隐私敏感 + 不愿数据上商业云",[99,3260,3261],{},"✅ 中国大陆开发者（Qwen \u002F DeepSeek 本地化）",[99,3263,3264],{},"❌ 非开发者 \u002F 不会 Python + Playwright",[99,3266,3267],{},"❌ 要 GUI \u002F 上手即用",[99,3269,3270],{},"❌ 生产级稳定（文档滞后 + 演进快）",[99,3272,3273],{},"❌ 团队协作 + 共享 workspace（用 Flowith \u002F Genspark Team）",[16,3275,1185],{"id":1185},[493,3277,3278,3284,3290],{},[99,3279,3280],{},[853,3281,3283],{"href":3282},"\u002Ftools\u002Fagent\u002Fplatform\u002Flangflow","Langflow 评测",[99,3285,3286],{},[853,3287,3289],{"href":3288},"\u002Ftools\u002Fagent\u002Fplatform\u002Fn8n","n8n 评测",[99,3291,3292],{},[853,3293,3295],{"href":3294},"\u002Ftools\u002Fagent\u002Fprotocol\u002Fcomposio","Composio 评测",[16,3297,1214],{"id":1214},[96,3299,3300,3307,3314,3321],{},[99,3301,3302,3303],{},"OpenManus GitHub 主仓库 + Foundation Agents 组织 ",[853,3304,3305],{"href":3305,"rel":3306},"https:\u002F\u002Fgithub.com\u002FFoundationAgents\u002FOpenManus",[857],[99,3308,3309,3310],{},"Foundation Agents — OpenManus 项目介绍 ",[853,3311,3312],{"href":3312,"rel":3313},"https:\u002F\u002Ffoundationagents.org\u002Fprojects\u002Fopenmanus\u002F",[857],[99,3315,3316,3317],{},"Toolsverse — OpenManus 评测 + 52k stars ",[853,3318,3319],{"href":3319,"rel":3320},"https:\u002F\u002Fthetoolsverse.com\u002Ftools\u002Fopenmanus",[857],[99,3322,3323,3324],{},"SoloSoft.dev — OpenManus 2026 Framework 综述 ",[853,3325,3326],{"href":3326,"rel":3327},"https:\u002F\u002Fwww.solosoft.dev\u002Fpost\u002Fopenmanus-agent-framework-2026\u002F",[857],[1259,3329,3330],{},"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 .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html pre.shiki code .sVt8B, html code.shiki .sVt8B{--shiki-default:#24292E;--shiki-dark:#E1E4E8}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":806,"searchDepth":1263,"depth":1263,"links":3332},[3333,3334,3335,3336,3337,3338,3339,3340,3341,3342],{"id":2116,"depth":1266,"text":2117},{"id":1548,"depth":1266,"text":1548},{"id":1775,"depth":1266,"text":1775},{"id":2804,"depth":1266,"text":2805},{"id":2286,"depth":1266,"text":2286},{"id":2354,"depth":1266,"text":2354},{"id":2496,"depth":1266,"text":2496},{"id":2542,"depth":1266,"text":2543},{"id":1185,"depth":1266,"text":1185},{"id":1214,"depth":1266,"text":1214},"\u002Fimg\u002Ftools\u002Fopenmanus.webp","OpenManus 真实评测：MetaGPT 核心成员 Xinbin Liang \u002F Jinyu Xiang \u002F Zhaoyang Yu \u002F Jiayi Zhang \u002F Sirui Hong 在 2025-03 推出的开源 Manus 替代方案，52,000+ GitHub stars。MIT 协议 + Python 模块化架构 + 多 agent orchestration（run_flow.py）+ Playwright 浏览器自动化 + MCP 工具支持 + DataAnalysis 内置模式 + OpenManus-RL 强化学习微调分支 + 自定义工具基类（BaseTool）。零邀请码、零订阅、零供应商绑定，唯一成本 = LLM API。",[3346,3349,3352,3355],{"q":3347,"a":3348},"OpenManus 和 Manus 是什么关系？","Manus 是商业 \u002F 邀请制的通用 AI agent 产品。OpenManus 是 MetaGPT 核心团队 2025-03 推出的开源复刻版，目标是『让所有人不靠邀请码就能用上类 Manus 能力』。功能覆盖：研究 \u002F 浏览器 \u002F 数据分析 \u002F 文件操作 \u002F 多步 reasoning。不是 Manus 官方出品。",{"q":3350,"a":3351},"和 LangChain \u002F AutoGPT \u002F CrewAI 怎么定位？","OpenManus 不是 framework，更像『可直接跑的通用 agent 实现』。LangChain 是 building block 框架；AutoGPT 是早期通用 agent；CrewAI 是多 agent 协作 framework。要『拉下来配 API 就能跑 Manus 风格任务』→ OpenManus；要『从底层搭自己的 agent』→ LangChain \u002F CrewAI；要『历史经典 + 学习』→ AutoGPT。OpenManus 内部用 LangChain-like 模块，可视为『现成实现』。",{"q":3353,"a":3354},"OpenManus-RL 是什么？","OpenManus 项目下的强化学习分支，提供 RL-based 微调方法优化 agent 性能。对研究 \u002F 高定制场景有价值，普通用户主仓库已经够用。",{"q":3356,"a":3357},"上手门槛？","需要 Python 3.12+ + 熟悉终端 + 自配 LLM API。无 GUI（虽然 web 监控界面在做）。documentation 偶尔滞后。非开发者建议先试 GUI 工具（Flowith \u002F Genspark），开发者 \u002F 研究者 + 想自托管 + 隐私敏感 → OpenManus。",[2038,3359,2649],"zh",{},"\u002Ftools\u002Fagent\u002Fgeneral\u002Fopenmanus",[2053,2054,2055,3363],"docker",[3365],{"plan":2791,"price":989,"features":3366,"notes":3367},"MIT 协议 + 全部功能 + 自托管 + 多 agent + 浏览器 + MCP + DataAnalysis + OpenManus-RL","LLM API 自付","MIT 完全免费开源 \u002F 用户自付 LLM API（GPT-4o \u002F Claude 3.5 \u002F Qwen VL Plus 任选）","2026-06-19",[3371],"onboarding\u002Fopen-source-general-agent",[2071,3373],"openmanus-deep-review",{"power":1686,"ux":1263,"price":1686,"cn_support":1674,"stability":1263},{"title":2686,"description":3344},[3377,3380,3382,3384],{"name":3378,"url":3305,"accessed":3379},"OpenManus GitHub（FoundationAgents 组织）","2026-06-24",{"name":3381,"url":3312,"accessed":3379},"Foundation Agents — OpenManus 项目介绍",{"name":3383,"url":3319,"accessed":3379},"Toolsverse — OpenManus 评测 + 52k stars",{"name":3385,"url":3326,"accessed":3379},"SoloSoft.dev — OpenManus 2026 Framework 综述","tools\u002Fagent\u002Fgeneral\u002Fopenmanus","MetaGPT 团队开源版 Manus——52k+ stars \u002F MIT \u002F 多 agent + 浏览器自动化 + MCP + DataAnalysis",[2101,3389,3390,3391,3392,3393],"multi-agent","browser-automation","mcp","metagpt","openmanus","想自托管复刻 Manus 全能 agent 体验 + 不愿等邀请码的开发者首选——浏览器 + 数据分析 + MCP 工具栈一站全。要 GUI \u002F 上手即用 \u002F 生产级稳定建议 Genspark \u002F Flowith 付费版。","bwDrZ3am1nVPoISuuo1SzArkD2h3Du6eKkMXUWgpesM",1784565436098]