Nox-Lumen MfgDeployment plans
Plan comparison
| Dimension | ☁️ SaaS | 🏢 On-premises | 🔀 Hybrid | 🇨🇳 Domestic stack |
|---|---|---|---|---|
| Data residency | Vendor cloud | Customer network | Apps on-prem + shared cloud | Customer + domestic suppliers |
| Code location | Vendor cloud | Customer DC | Customer DC | Customer DC |
| LLM | Managed channels | Self-hosted / intranet API | Mix | Domestic models on-prem |
| Hardware | None | Customer supplies | Partially customer | Customer (domestic CPU/GPU) |
| Operations | Vendor managed | Customer (+ optional MSP) | Mixed | Customer |
| Licensing | Usage subscription | Non-commercial source license | Bundle quote | Non-commercial source license |
| Ideal for | Quick trials · small teams | Large enterprises · regulated data | Cross-border sovereignty tension | Defense / government / classified |
SaaS
Fits when
- Fastest time-to-value; no ops headcount
- Cloud processing acceptable under policy
- Teams ~10–100 seats
Delivered
- Tenant workspace
- Seat or usage-based billing
- Vendor owns updates, observability, backups, incident response
- Marketplace skills included
Constraints
- Model spend aggregated at platform rates
- Source code not shipped
- Cross-border customers must validate legal posture
On-premises (preferred for large regulated customers)
Fits when
- Data must not leave the LAN
- Heavy customization
- ≥3 year strategic commitment
Delivered
- Full source (web + API + agent runtime + skill library)
- Deployment assets (Compose / Helm charts)
- Runbooks + admin training
- First-year support (quarterly releases + defect fixes)
- Project-scoped custom skills as contracted
Hardware sketch
| Scale | Topology | Suggested spec |
|---|---|---|
| Small (under 50 seats) | Single stack | 16 vCPU / 64 GB / 2 TB SSD |
| Medium (50–500) | 3 app + 2 data nodes | 16 / 64 / 1 TB per node |
| Large (>500) | 5+ app, 3+ data, 5+ workers | capacity-modeled |
Self-hosted LLM inference: plan 1–8× A100 / H800-class (or domestic accelerators) depending on model size.
Licensing highlights (non-commercial template)
- ✅ Internal enterprise use
- ✅ Distribution to subsidiaries / internal groups
- ✅ Forking / customization
- ❌ External resale
- ❌ Rebadged public SaaS offerings
Contract governs final terms.
Hybrid
Fits when
- Multi-region sovereignty requirements
- Sensitive core + cloud-acceptable peripheral services
- Lower ops burden without surrendering crown-jewel data
Example split
| Piece | Placement |
|---|---|
| Agent runtime + API | Customer network |
| KB + session stores | Customer network |
| LLM | Customer-hosted or compliant cloud channel |
| Object storage | Customer MinIO |
| Observability logs | Vendor cloud (sanitized) |
Domestic technology stack
Fits when
- Defense, government, regulated critical infrastructure
- End-to-end localization mandates
- Controls: classified level-3, crypto evaluation, domestic certification
Typical stack
| Layer | Options |
|---|---|
| CPU | Phytium / Loongson / Kunpeng / Zhaoxin |
| OS | Unity UOS / Kylin / openEuler |
| Database | Dameng / Kingbase / OceanBase |
| Middleware | Domestic Redis-class / GaussDB variants |
| LLM | DeepSeek / Qwen / Zhipu / ERNIE / Spark |
| Accelerators | Ascend / Cambricon / Biren |
Compliance support
- Playbooks for classified deployments
- Cryptographic evaluation facilitation
- Compatibility statements for domestic procurement catalogs
Decision tree
Rendering diagram…