ponshu-room-lite/docs/architecture/archive/MASTER_PROMPT_FOR_AI.md

39 lines
1.8 KiB
Markdown

# PROMPT FOR AI: Project "Posimai" Architecture Context (Synology VMM Edition)
<!-- Copy the content below this line and paste it to Claude or ChatGPT -->
# Project Context: "Posimai" Digital Factory (Local VMM)
This prompt establishes the architectural context for the "Posimai" project.
We have decided on the **"Synology VMM (Virtual Machine Manager)"** architecture.
## 1. Core Architecture: "The Zero-Cost Local Factory"
We are leveraging the user's high-spec NAS (16GB RAM) to host the entire stack locally.
| Layer | Technology | Role |
| :--- | :--- | :--- |
| **Hardware** | **Synology NAS (16GB RAM)** | The physical host. |
| **Virtualization** | **Synology VMM** | Running an **Ubuntu Server VM** inside the NAS. |
| **Control Plane** | **Dokploy** (inside VM) | CI/CD Engine. Manages containers inside the VM. |
| **App Runtime** | **Docker** (inside VM) | Hosting Sake/Incense App APIs. |
| **Data Plane** | **PostgreSQL** (on DSM/Docker) | Master Data, accessible from VM via local IP. |
| **Network** | **Tailscale** | Exposing the VM services securely (Funnel/VPN). |
## 2. Resource Split & AI Location
* **Host (DSM): 12GB** for Database, Immich (CLIP), Ollama (Nightly).
* **Guest (VM): 4GB** for Dokploy & Apps.
* **Cloud: Google Cloud** for Gemini 2.5 (Real-time Vision).
## 3. Optimization & Safety Rules
* **Fallback**: If Cloud is down, use **Ollama** locally for low-res analysis.
* **Scheduling**: Heavy tasks (Immich scan) must run at **3 AM**.
* **Safety**: Google Cloud Quota limits MUST be set to prevent billing runaway.
* **Caching**: Future goal is Vector Search to reuse Gemini results.
## 4. Product Roadmap
* **Product A**: **Sake Note** (Existing)
* **Product B**: **Incense Note** (Upcoming)
* **Posimai Core**: Shared Flutter Library.
---
<!-- End of Prompt -->