Hermes CI FixandHermes AI
e1a9b25afa
- nginx 安全加固 (CSP, HSTS, 缓存策略) - 共享 style.css - 138个页面全部接入 Co-authored-by: Hermes AI <agent@hermes>
211 lines
13 KiB
HTML
211 lines
13 KiB
HTML
<!DOCTYPE html>
|
||
<html lang="zh-CN">
|
||
<head>
|
||
<meta charset="UTF-8">
|
||
<meta name="viewport" content="width=device-width, initial-scale=1.0">
|
||
<title>博海科技 · AI Agent框架选型评估报告</title>
|
||
<style>
|
||
:root {
|
||
--bg: #0f172a; --card: #1e293b; --border: #334155;
|
||
--text: #e2e8f0; --muted: #94a3b8; --accent: #38bdf8;
|
||
--green: #22c55e; --yellow: #eab308; --red: #ef4444;
|
||
--gold: #f59e0b;
|
||
}
|
||
* { margin: 0; padding: 0; box-sizing: border-box; }
|
||
body { font-family: -apple-system, 'Segoe UI', 'Noto Sans SC', sans-serif; background: var(--bg); color: var(--text); line-height: 1.7; padding: 2rem 1rem; }
|
||
.container { max-width: 1000px; margin: 0 auto; }
|
||
h1 { font-size: 1.8rem; border-bottom: 2px solid var(--accent); padding-bottom: .5rem; margin-bottom: 1rem; }
|
||
h2 { font-size: 1.3rem; color: var(--accent); margin: 2rem 0 .8rem; }
|
||
h3 { font-size: 1.1rem; color: var(--gold); margin: 1.5rem 0 .5rem; }
|
||
.meta { color: var(--muted); font-size: .85rem; margin-bottom: 2rem; }
|
||
p, li { color: var(--text); margin-bottom: .5rem; }
|
||
ul, ol { padding-left: 1.5rem; margin: .5rem 0 1rem; }
|
||
table { width: 100%; border-collapse: collapse; margin: 1rem 0; font-size: .9rem; }
|
||
th, td { border: 1px solid var(--border); padding: .5rem .6rem; text-align: left; }
|
||
th { background: var(--card); color: var(--accent); font-weight: 600; white-space: nowrap; }
|
||
.score-5 { color: var(--green); font-weight: bold; }
|
||
.score-4 { color: #4ade80; }
|
||
.score-3 { color: var(--yellow); }
|
||
.score-2 { color: #fb923c; }
|
||
.score-1 { color: var(--red); }
|
||
|
||
.rank-card { background: var(--card); border: 1px solid var(--border); border-radius: 8px; padding: 1rem; margin: .8rem 0; }
|
||
.rank-card .title { font-size: 1.1rem; font-weight: 600; margin-bottom: .3rem; }
|
||
.rank-card .score { font-size: 2rem; font-weight: 700; }
|
||
.rank-card .verdict { margin-top: .3rem; font-size: .9rem; color: var(--muted); }
|
||
|
||
.recommend { background: linear-gradient(135deg, #1e3a5f, #1e293b); border: 1px solid var(--accent); border-radius: 10px; padding: 1.2rem; margin: 1.5rem 0; }
|
||
.recommend h3 { margin-top: 0; color: var(--accent); }
|
||
.recommend ul { margin-bottom: 0; }
|
||
|
||
.badge { display: inline-block; padding: .15rem .5rem; border-radius: 4px; font-size: .75rem; font-weight: 600; }
|
||
.badge-green { background: #14532d; color: var(--green); }
|
||
.badge-yellow { background: #422006; color: var(--yellow); }
|
||
.badge-red { background: #450a0a; color: var(--red); }
|
||
.badge-blue { background: #0c4a6e; color: var(--accent); }
|
||
|
||
.pattern-item { background: var(--card); border-left: 3px solid var(--yellow); padding: .8rem 1rem; margin: .6rem 0; border-radius: 0 6px 6px 0; }
|
||
.pattern-item h4 { color: var(--gold); margin-bottom: .3rem; }
|
||
.pattern-item p { font-size: .9rem; color: var(--muted); margin-bottom: 0; }
|
||
|
||
@media (max-width: 700px) { table { font-size: .75rem; } th, td { padding: .3rem .4rem; } }
|
||
</style>
|
||
<link rel="stylesheet" href="/style.css">
|
||
</head>
|
||
<body>
|
||
<div class="container">
|
||
|
||
<h1>博海科技 · AI Agent框架选型评估报告</h1>
|
||
<div class="meta">
|
||
评估日期: 2026-07-10 | 方法论来源: 12维度系统化选型 | 技术栈基线: Hermes Agent 0.18.0 / 13 Profiles / Docker / Prometheus+Grafana+Loki
|
||
</div>
|
||
|
||
<h2>一、评估背景</h2>
|
||
<p>博海科技当前运行 <strong>13个AI Agent Profiles</strong>（运维/财务/项目/研学/行政/商务/品宣/营销/客服/员工通道/老板通道等），基于 <strong>Hermes Agent 0.18.0</strong> 框架。随着企业AI Agents市场以49.6% CAGR增长（2033年预计达1829.7亿美元），需评估是否引入或替换框架以支撑下一阶段发展。</p>
|
||
|
||
<h2>二、评分矩阵</h2>
|
||
<table>
|
||
<tr>
|
||
<th>框架</th><th>部署模式</th><th>语言接口</th><th>多Agent</th><th>工作流</th><th>工具生态</th>
|
||
<th>模型支持</th><th>RAG</th><th>记忆状态</th><th>生产运维</th><th>安全权限</th><th>成本</th><th>社区</th>
|
||
<th>总分</th>
|
||
</tr>
|
||
<tr>
|
||
<td><strong>LangChain</strong></td>
|
||
<td class="score-4">4</td><td class="score-5">5</td><td class="score-2">2</td><td class="score-4">4</td><td class="score-5">5</td>
|
||
<td class="score-4">4</td><td class="score-5">5</td><td class="score-3">3</td><td class="score-4">4</td><td class="score-2">2</td>
|
||
<td class="score-4">4</td><td class="score-5">5</td><td class="score-5">47</td>
|
||
</tr>
|
||
<tr>
|
||
<td><strong>Dify</strong></td>
|
||
<td class="score-5">5</td><td class="score-3">3</td><td class="score-3">3</td><td class="score-5">5</td><td class="score-4">4</td>
|
||
<td class="score-4">4</td><td class="score-5">5</td><td class="score-3">3</td><td class="score-3">3</td><td class="score-4">4</td>
|
||
<td class="score-4">4</td><td class="score-4">4</td><td class="score-5">47</td>
|
||
</tr>
|
||
<tr>
|
||
<td><strong>LangGraph</strong></td>
|
||
<td class="score-4">4</td><td class="score-4">4</td><td class="score-4">4</td><td class="score-5">5</td><td class="score-3">3</td>
|
||
<td class="score-4">4</td><td class="score-3">3</td><td class="score-5">5</td><td class="score-4">4</td><td class="score-3">3</td>
|
||
<td class="score-3">3</td><td class="score-4">4</td><td class="score-5">46</td>
|
||
</tr>
|
||
<tr>
|
||
<td><strong>CrewAI</strong></td>
|
||
<td class="score-4">4</td><td class="score-4">4</td><td class="score-5">5</td><td class="score-3">3</td><td class="score-2">2</td>
|
||
<td class="score-4">4</td><td class="score-2">2</td><td class="score-3">3</td><td class="score-2">2</td><td class="score-2">2</td>
|
||
<td class="score-5">5</td><td class="score-3">3</td><td class="score-4">39</td>
|
||
</tr>
|
||
<tr>
|
||
<td><strong>AutoGen</strong></td>
|
||
<td class="score-3">3</td><td class="score-3">3</td><td class="score-5">5</td><td class="score-3">3</td><td class="score-3">3</td>
|
||
<td class="score-3">3</td><td class="score-2">2</td><td class="score-3">3</td><td class="score-2">2</td><td class="score-2">2</td>
|
||
<td class="score-4">4</td><td class="score-3">3</td><td class="score-4">36</td>
|
||
</tr>
|
||
<tr>
|
||
<td><strong>Coze(扣子)</strong></td>
|
||
<td class="score-1">1</td><td class="score-2">2</td><td class="score-2">2</td><td class="score-4">4</td><td class="score-3">3</td>
|
||
<td class="score-3">3</td><td class="score-3">3</td><td class="score-2">2</td><td class="score-2">2</td><td class="score-2">2</td>
|
||
<td class="score-3">3</td><td class="score-3">3</td><td class="score-2">30</td>
|
||
</tr>
|
||
</table>
|
||
|
||
<h2>三、排名与解读</h2>
|
||
|
||
<div class="rank-card">
|
||
<div class="title">🥇 LangChain — 47/60 <span class="badge badge-blue">生态之王</span></div>
|
||
<div class="score score-5">★★★★★</div>
|
||
<div class="verdict">优势：工具生态无可匹敌（11.8万Star）、RAG最成熟、LangSmith与我们的Grafana互补。短板：单Agent为主，安全权限需自建。适合作为我们Hermes的外部工具链补充。</div>
|
||
</div>
|
||
|
||
<div class="rank-card">
|
||
<div class="title">🥇 Dify — 47/60 <span class="badge badge-green">自托管最优</span></div>
|
||
<div class="score score-5">★★★★★</div>
|
||
<div class="verdict">优势：Docker一键部署（最友好）、可视化编排降低门槛、内置RAG全流程、中文社区活跃。短板：多Agent协作弱。适合作为企业知识库/RAG场景的独立AI应用平台。</div>
|
||
</div>
|
||
|
||
<div class="rank-card">
|
||
<div class="title">🥉 LangGraph — 46/60 <span class="badge badge-yellow">生产级编排</span></div>
|
||
<div class="score score-4">★★★★☆</div>
|
||
<div class="verdict">优势：图结构编排最精细、状态持久化/checkpoint最完善、Human-in-the-loop原生支持。短板：学习曲线陡、Platform需要付费。最适合我们复杂多步Agent任务的编排增强。</div>
|
||
</div>
|
||
|
||
<div class="rank-card">
|
||
<div class="title">4. CrewAI — 39/60</div>
|
||
<div class="score score-3">★★★☆☆</div>
|
||
<div class="verdict">多Agent角色化协作的理念与我们的Profile模式高度一致，但生产级能力欠缺。如果我们要扩展多Agent协作，可以借鉴其Role/Agent/Task三层抽象模式，而非直接迁移。</div>
|
||
</div>
|
||
|
||
<div class="rank-card">
|
||
<div class="title">5. AutoGen — 36/60</div>
|
||
<div class="score score-3">★★★☆☆</div>
|
||
<div class="verdict">微软背书，多Agent对话式协作有特色，但偏研究风格，运维体系弱，与我们的生产环境要求有差距。</div>
|
||
</div>
|
||
|
||
<div class="rank-card">
|
||
<div class="title">6. Coze(扣子) — 30/60</div>
|
||
<div class="score score-2">★★☆☆☆</div>
|
||
<div class="verdict">云托管封闭平台，无法自托管Docker化，数据在字节服务器。与博海「自托管优先」战略根本冲突，不做考虑。</div>
|
||
</div>
|
||
|
||
<h2>四、核心结论：战略定位</h2>
|
||
|
||
<div class="recommend">
|
||
<h3>🚀 推荐方案：坚守Hermes + LangGraph补充，不迁不换</h3>
|
||
<ul>
|
||
<li><strong>核心框架不动</strong>：Hermes Agent 0.18.0 已深度适配我们的13个Profile架构、系统总线（:8660）、sandbox-runner、approval-gate等基础设施。迁移成本极高，无必要。</li>
|
||
<li><strong>引入LangGraph作为编排增强</strong>：针对需要复杂多步状态管理的Agent任务（如长周期运维审计、跨Bot协作流程），引入LangGraph的图编排能力，通过Python SDK解析执行。</li>
|
||
<li><strong>Dify作为RAG/KB独立平台</strong>：如果企业知识库/RAG需求增长，Dify Docker Compose一键部署，可以快速搭建独立的AI问答平台，与现有Hermes架构互补。</li>
|
||
<li><strong>借鉴CrewAI的角色模式</strong>：CrewAI的Role/Agent/Task三层抽象设计理念已体现在我们的Profile+SOUL.md体系，持续优化即可。</li>
|
||
</ul>
|
||
</div>
|
||
|
||
<h2>五、本次评估的可落地模式提取</h2>
|
||
|
||
<p>从六大框架中提取<span class="badge badge-yellow">3个可复用模式</span>，已排好优先级：</p>
|
||
|
||
<div class="pattern-item">
|
||
<h4>🔄 模式A：Checkpoint/State持久化（LangGraph ✕ Hermes）</h4>
|
||
<p>LangGraph的checkpoint机制支持Agent执行到任意节点保存/恢复状态。Hermes目前缺乏任务级状态持久化——长任务中断后从头开始。建议在Hermes中实现任务快照（checkpoint），避免长周期Agent操作因超时或重启丢失进度。<br><span class="badge badge-yellow">优先级：高</span></p>
|
||
</div>
|
||
|
||
<div class="pattern-item">
|
||
<h4>🔍 模式B：分布式Tracing/Observability（LangSmith ↔ Prometheus）</h4>
|
||
<p>LangSmith的LangChain调用链追踪能力，结合我们的Grafana/Prometheus/Loki体系，可以构建完整的「Agent调用链」可观测性。每个Profile的每轮Agent执行都打上trace_id，实现全链路追踪。<br><span class="badge badge-yellow">优先级：中</span></p>
|
||
</div>
|
||
|
||
<div class="pattern-item">
|
||
<h4>📋 模式C：内置RBAC权限模型（Dify Security → Bot安全治理）</h4>
|
||
<p>Dify内置RBAC/API Key管理/审计日志，这对我们多个Profile的权限隔离有借鉴意义。当前我们通过approval-gate + website_blocklist + command_allowlist三层安全，可进一步补充「Profile间操作审计」能力。<br><span class="badge badge-yellow">优先级：低（现有体系够用）</span></p>
|
||
</div>
|
||
|
||
<h2>六、行动建议</h2>
|
||
<ul>
|
||
<li><strong>✅ 立即做</strong>：将本评估报告归档到 agent-infrastructure-patterns 技能实战记录</li>
|
||
<li><strong>🔜 两周内</strong>：评估LangGraph checkpoint模式与Hermes的集成方案（模式A）</li>
|
||
<li><strong>📅 按需</strong>：当RAG/知识库需求增长时，启动Dify试点部署（已有17个Docker容器，再加一个无压力）</li>
|
||
</ul>
|
||
|
||
<h2>七、对标总结：Hermes Agent vs 六大框架</h2>
|
||
|
||
<table>
|
||
<tr><th>维度</th><th>Hermes Agent（我们）</th><th>对标框架</th><th>差距/优势</th></tr>
|
||
<tr><td>多Agent协作</td><td>✅ 13 Profiles + System-Bus</td><td>CrewAI / AutoGen</td><td><span class="badge badge-green">优势：已落地，比他们更早用于生产</span></td></tr>
|
||
<tr><td>自托管/部署</td><td>✅ Docker化，API Server 127.0.0.1</td><td>Dify / LangChain</td><td><span class="badge badge-green">平齐，可深度定制</span></td></tr>
|
||
<tr><td>工具/插件生态</td><td>⚠️ 自定义工具为主，无市场</td><td>LangChain (11.8万Star)</td><td><span class="badge badge-yellow">差距：工具丰富度远不如</span></td></tr>
|
||
<tr><td>可视化编排</td><td>❌ 纯代码配置</td><td>Dify / Coze</td><td><span class="badge badge-yellow">差距：非技术成员需要学习</span></td></tr>
|
||
<tr><td>生产级监控</td><td>✅ Prometheus+Grafana+Loki</td><td>LangSmith</td><td><span class="badge badge-green">优势：通用监控方案更成熟</span></td></tr>
|
||
<tr><td>安全治理</td><td>✅ approval-gate+tirith+allowlist</td><td>Dify RBAC</td><td><span class="badge badge-green">优势：纵深防御体系更强</span></td></tr>
|
||
<tr><td>RAG/知识管理</td><td>⚠️ Wiki.js + Loki日志检索</td><td>Dify (内置RAG)</td><td><span class="badge badge-yellow">差距：缺乏结构化RAG管道</span></td></tr>
|
||
<tr><td>社区/生态</td><td>❌ Hermes社区较小</td><td>LangChain 11.8万Star</td><td><span class="badge badge-yellow">差距：遇到问题靠自己排查</span></td></tr>
|
||
</table>
|
||
|
||
<hr style="border-color: var(--border); margin: 2rem 0;">
|
||
|
||
<p style="text-align:center; color:var(--muted); font-size:.85rem;">
|
||
博海科技 · 运维BOT · 2026-07-10<br>
|
||
基于《AI Agent开发框架怎么选？12维度系统化选型方法论》(奇崛数字传媒科技) 定制评估
|
||
</p>
|
||
|
||
</div>
|
||
</body>
|
||
</html>
|