Vertai seamlessly bridges on-premise local models and cloud intelligence — giving your entire organization access to Openclaw & Hermes Agent at a fraction of the cost, with zero data leakage risk. 垂域智能一体机无缝切换内外网模型，让全公司以极低成本安全使用 Openclaw 和 Hermes Agent，彻底消除数据泄漏风险。
A purpose-built rack server pre-loaded with local LLMs, our SmartRouter middleware, and Vertai's desensitization engine — deployable on-premise in under 2 hours. 专为企业设计的机架服务器，预装本地大模型、SmartRouter 智能路由中间件及脱敏引擎，2 小时内即可完成本地化部署。
Vertai ships with Qwen 3.5-27B Dense pre-loaded and optimized. Your team gets instant, private AI for everyday tasks — no API costs, no data leaving the building. When harder tasks need cloud-grade models, SmartRouter routes seamlessly and our desensitization engine strips all sensitive data first. 垂域一体机预装并优化了 Qwen 3.5-27B Dense 模型，团队日常任务可即时获得私有 AI 支持，无 API 费用，数据不出内网。面对复杂任务需调用云端模型时，SmartRouter 无缝路由，脱敏引擎在发出请求前自动去除所有敏感信息。
Zero-leakage guarantee — sensitive data never reaches external APIs without desensitization 零泄漏保证 — 敏感数据未经脱敏绝不到达外部 API
Instant deployment — rack mount, power on, deploy in under 2 hours with zero cloud dependency 快速部署 — 上架通电，2 小时内完成部署，零云端依赖
Enterprise-wide access — support 500+ concurrent users at costs 86% lower than SaaS alternatives 全员覆盖 — 支持 500+ 并发用户，成本较 SaaS 方案低 86%
SmartRouter automatically analyzes every query and routes it to the optimal model — simple tasks stay local (free & fast), complex reasoning routes to cloud models. You get the best intelligence at the lowest possible cost, with zero manual configuration. SmartRouter 自动分析每条查询并路由至最优模型 —— 简单任务留在本地（免费且快速），复杂推理路由至云端大模型。无需手动配置，以最低成本获得最佳智能。
Before any data reaches external cloud models, our real-time pipeline detects and masks PII, financial records, IP addresses, and proprietary content. After the cloud response arrives, sensitive placeholders are restored in full context — seamlessly. 在任何数据到达外部云端模型之前，实时管道自动检测并屏蔽个人信息、财务记录、IP 地址及专有内容。云端响应返回后，敏感占位符在完整上下文中无缝还原。
Every request is authenticated and authorized. No data leaves the appliance without explicit routing rules and desensitization checks.每次请求均经过身份验证与授权，未经路由规则和脱敏检查，数据绝不离开一体机。
One OpenAI-compatible endpoint for your entire organization. No app rewrites needed — just point to Vertai and everything works.一个 OpenAI 兼容端点服务全公司，无需修改应用代码，直接指向垂域智能即可。
Manage Qwen 3.5, Openclaw, and Hermes Agent from a single control plane with fine-grained per-department usage policies.从统一控制面管理 Qwen 3.5、Openclaw 和 Hermes Agent，支持按部门精细化配置使用策略。
Every query, routing decision, and desensitization event is logged immutably. Supports GDPR, PIPL, and SOC2 compliance workflows.所有查询、路由决策和脱敏事件不可篡改地记录，支持 GDPR、个人信息保护法及 SOC2 合规流程。
Real-time breakdown of local vs cloud spend by team, user, and task type. Set budgets and auto-downgrade routing when limits approach.按团队、用户和任务类型实时呈现本地与云端费用，可设置预算并在临近上限时自动降级路由。
Automatic failover between local and cloud models. If the intranet model is under load, SmartRouter seamlessly routes to cloud with desensitization enforced.本地与云端模型之间自动故障切换，内网模型负载过高时，SmartRouter 无缝路由至云端并强制执行脱敏。
A layered architecture that keeps your data safe at every step — from intranet query to cloud response and back. 分层架构在每个步骤保护数据安全 —— 从内网查询到云端响应全程可控。
Pre-loaded and optimized for your hardware. Handles 80% of enterprise queries locally — Q&A, drafting, translation, summarization — with sub-50ms latency and zero cost per query.预装并针对您的硬件优化，在本地处理 80% 的企业查询 —— 问答、起草、翻译、摘要 —— 延迟低于 50ms，每次查询零成本。
State-of-the-art reasoning and code generation. Routed for medium-to-hard tasks by SmartRouter, always behind the Data Shield desensitization layer before any request leaves the intranet.顶尖推理与代码生成能力。由 SmartRouter 路由至中高难度任务，所有请求在离开内网前均经过数据盾脱敏层保护。
Multi-step agentic reasoning with tool use, RAG, and long-horizon planning. Powers complex enterprise workflows — financial analysis, legal review, research synthesis — with full desensitization enforced.多步智能体推理，支持工具调用、RAG 与长程规划，驱动复杂企业工作流 —— 财务分析、法律审查、研究综合 —— 全程强制脱敏。
One-time hardware investment + annual software subscription. No per-query costs for local model usage — ever. 一次性硬件投资 + 年度软件订阅。本地模型使用永不收取每次查询费用。
|
All plans include SmartRouter™, Data Shield™, Openclaw & Hermes Agent access.
所有方案均包含 SmartRouter™、Data Shield™ 及 Openclaw / Hermes Agent 接入。
|
Standard
$2,399¥49,800
one-time · hardware一次性 · 硬件
20–70B parameter models20–70B 参数量模型
|
Most Popular最受欢迎
Pro
$4,899¥99,800
one-time · hardware一次性 · 硬件
100B+ quantized models100B+ 量化模型
|
Enterprise
$220K – 350K¥150–250万
one-time · hardware一次性 · 硬件
Full-weight frontier models全量前沿模型
|
|---|---|---|---|
| Hardware硬件配置 | Unified memory architecture统一内存架构 | Discrete GPU architecture独立 GPU 架构 | Discrete GPU architecture独立 GPU 架构 |
| GPU / Unified Memory统一内存Available memory for inference可用推理内存 | 128 GB Unified shared memory统一共享内存 No discrete GPU needed无需独立显卡 | 320 GB VRAM · 4× discrete GPU | 640 GB VRAM · 8× H100 80G |
| CPU / System RAM系统内存Processor & memory处理器与内存 | High-perf SoC高性能 SoC 一体芯片 Memory shared with compute内存与算力共享 | Dual Xeon Gold512 GB DDR5 ECC | Dual Xeon Platinum2 TB DDR5 ECC |
| Storage存储NVMe SSD | 4 TB NVMe | 16 TB NVMe | 64 TB NVMe RAID |
| Model Capability模型能力 | |||
| Supported Models可运行模型Local inference本地推理 | Qwen2.5-32B Llama3-70B-Q4 DeepSeek-R1-32B 20–70B quantized20–70B 量化模型 | Qwen2.5-72B DeepSeek-R1-70B Llama3.1-405B-Q4 100B+ quantized models100B+ 量化模型 | Kimi 2.6 GLM 5.1 Qwen 3.5 Full Full-weight frontier models全量前沿旗舰模型 |
| Model Parameters模型参数量Typical range典型区间 | 20 – 70 BDense or Q4/Q8 quant稠密或 Q4/Q8 量化 | 70 – 200 B+Q4/Q8 quantizedQ4/Q8 量化 | 600 B+Full-weight, no compression全量，不压缩 |
| Throughput吞吐性能 | |||
| Token Speed生成速度Output tokens/sec输出 tokens/秒 | ~30 tok/sat 32B model load32B 模型下 | ~80 tok/sat 72B Q4 model load72B Q4 模型下 | ~120 tok/sat full-weight Qwen 3.5Qwen 3.5 全量模型下 |
| Context Window上下文窗口Max tokens per session单会话最大 tokens | 128K – 256K | 128K – 256K | 256K – 1M |
| Concurrency并发能力 | |||
| Concurrent Connections并发连接数Simultaneous requests同时处理请求数 | 15concurrent sessions并发会话 | 50concurrent sessions并发会话 | 200+concurrent sessions并发会话 |
| Parallel Agents并行 Agent 数量Simultaneous Hermes agents同时运行的 Hermes Agent | 5parallel agents并行智能体 | 20parallel agents并行智能体 | 100+parallel agents并行智能体 |
| Registered Users注册用户上限Total platform users平台总账户数 | 100 users | 500 users | 无上限 unlimited |
| Software & Support软件与服务 | |||
| SmartRouter™Intelligent routing智能路由 | ✓ | ✓ + Custom policies+ 自定义路由策略 | ✓ Full API + custom rules完整 API + 定制规则 |
| Data Shield™Auto-desensitization自动脱敏 | ✓ | ✓ | ✓ Custom rule engine定制脱敏规则引擎 |
| Support技术支持 | Email · 5×8邮件 · 5×8 | Priority · 7×24 + SLA优先 · 7×24 + SLA | Dedicated CSM · On-site专属客户经理 · 驻场 |
| * Annual software subscription sold separately.* 年度软件授权费用另计。 | Contact Sales联系销售 | Request Demo →预约演示 → | Get a Quote获取报价 |
We're a team of engineers, researchers, and enterprise builders working on one of the most important infrastructure challenges in AI. Join us. 我们是一支由工程师、研究员和企业构建者组成的团队，正在攻克 AI 领域最重要的基础设施挑战之一。欢迎加入。
Our enterprise sales team responds within 24 hours. For existing customers, reach support directly through your dashboard. 我们的企业销售团队将在 24 小时内回复。现有客户可直接通过控制台联系支持团队。