|
| 1 | +# LifeOS Infrastructure Bill of Materials |
| 2 | + |
| 3 | +## Overview |
| 4 | + |
| 5 | +This guide provides concrete, budget-tiered hardware recommendations for self-hosted LifeOS deployments. These BOMs assume: |
| 6 | + |
| 7 | +- Self-hosted LifeOS with always-on web/API/database/sync services |
| 8 | +- Mobile clients via Expo EAS Build (hosted service for React Native binaries—no local Mac required) |
| 9 | +- Proper backup discipline and UPS protection |
| 10 | +- Practical local AI / privacy work where cost-justified |
| 11 | + |
| 12 | +Prices are representative list prices (B&H, vendor MSRPs) as of 2026. Adjust by region and current market. |
| 13 | + |
| 14 | +--- |
| 15 | + |
| 16 | +## Budget Build — One-Box Starter |
| 17 | + |
| 18 | +**Best for:** 1 user or a couple, normal always-on LifeOS services, cloud AI, no serious local LLM work. |
| 19 | + |
| 20 | +### Hardware |
| 21 | + |
| 22 | +| Component | Model | Notes | Price | |
| 23 | +|-----------|-------|-------|-------| |
| 24 | +| **Primary Host** | Beelink EQR6 (AMD Ryzen 5 6600U, 24GB LPDDR5, 500GB SSD) | x86 always-on, dual M.2 PCIe 4.0 up to 8TB, 85W PSU | $509.00 | |
| 25 | +| **UPS** | APC BX1500M (1500VA / 900W, 10 outlets) | Protects against brief power loss | $189.99 | |
| 26 | +| **Backup Drive** | WD Elements Desktop 8TB (external HDD) | Offline rotation backup | $209.99 | |
| 27 | +| | | **Estimated Total** | **$908.98** | |
| 28 | + |
| 29 | +### Why This Tier |
| 30 | + |
| 31 | +This is the **cheapest build I'd still call "real"**. It gives you: |
| 32 | + |
| 33 | +- A dedicated, always-on x86 box for app/API/database workloads |
| 34 | +- Enough RAM for typical LifeOS services (sync, calendar, notes, basic orchestration) |
| 35 | +- **No local LLM burden:** cloud AI or Ollama on small models only |
| 36 | +- UPS protection against brownouts and brief outages |
| 37 | +- Separate backup disk for rotation off-site |
| 38 | + |
| 39 | +Cost savings come from: |
| 40 | +- Single mini PC (no NAS) |
| 41 | +- Bare UPS (no redundancy layer) |
| 42 | +- External HDD backup instead of NAS RAID |
| 43 | + |
| 44 | +### Typical Deployment |
| 45 | + |
| 46 | +- LifeOS core (web, API, database, NATS, module-loader) runs on the Beelink |
| 47 | +- Backup drops to external USB disk on a schedule |
| 48 | +- Whisper/voice runs on local CPU (acceptable ~1s latency) |
| 49 | +- Larger models or heavy inference → cloud LLM API |
| 50 | + |
| 51 | +--- |
| 52 | + |
| 53 | +## Recommended Build — Serious Home Production Node |
| 54 | + |
| 55 | +**Best for:** A real Personal Operations OS / early Household Coordination OS deployment with proper local storage, backup discipline, and room to grow. |
| 56 | + |
| 57 | +### Hardware |
| 58 | + |
| 59 | +| Component | Model | Notes | Price | |
| 60 | +|-----------|-------|-------|-------| |
| 61 | +| **Primary Host** | MINISFORUM UM890 Pro (AMD Ryzen 9 8945HS, 64GB RAM, 1TB SSD) | Dual 2.5GbE, dual M.2 PCIe 4.0, Oculink, USB4, up to 96GB RAM | $1,135.00 | |
| 62 | +| **NAS / Online Backup** | Synology DS923+ (4-bay NAS) | Always-on, 10GbE & NVMe cache support, `>50TB` potential | $599.99 | |
| 63 | +| **NAS Drives** | 2 × WD Red Plus 8TB (B&H 2-pack) | RAID-optimized for NAS up to 8 bays | $399.98 | |
| 64 | +| **UPS** | APC BX1500M (1500VA / 900W) | Protects NAS + primary host | $189.99 | |
| 65 | +| **Off-Box Rotation Backup** | WD Elements Desktop 8TB (external HDD) | Rotate off-site monthly/quarterly | $209.99 | |
| 66 | +| | | **Estimated Total** | **$2,534.95** | |
| 67 | + |
| 68 | +### Why This Is the Sweet Spot |
| 69 | + |
| 70 | +This is **the tier I would actually buy first for LifeOS**: |
| 71 | + |
| 72 | +- The mini PC is strong enough for app/API/database/worker orchestration without choking |
| 73 | +- The DS923+ provides an always-on, managed backup and file target with RAID discipline |
| 74 | +- The external disk gives you a second copy you can rotate off-box for disaster recovery |
| 75 | +- Path to upgrade later: swap in a 10GbE card or M.2 NVMe cache on the NAS without replacing hardware |
| 76 | + |
| 77 | +### Privacy / Local Voice Note |
| 78 | + |
| 79 | +**Important nuance:** If your privacy goal is **local voice but not full local LLM inference**, this tier is often enough. |
| 80 | + |
| 81 | +Home Assistant's local voice docs report: |
| 82 | +- Whisper on a Raspberry Pi 4: ~8 seconds per request |
| 83 | +- Whisper on an Intel NUC: `<1 second` |
| 84 | + |
| 85 | +This is a good signal that **local voice is much lighter than serious local model serving**. You can run private transcription on the UM890 without needing the Privacy-max tier. |
| 86 | + |
| 87 | +### Typical Deployment |
| 88 | + |
| 89 | +- LifeOS core on the MINISFORUM (64GB supports medium-scale agent orchestration) |
| 90 | +- DS923+ runs Synology services (backup, media, optional Synology Apps) |
| 91 | +- 2× WD Red in RAID 1 (mirrored, protected against single disk failure) |
| 92 | +- External WD Elements rotates off-site, keeping a copy safe from local disaster |
| 93 | +- Local Whisper for always-private voice (CPU-bound on the UM890) |
| 94 | +- Cloud LLM or small Ollama models for heavier inference |
| 95 | + |
| 96 | +--- |
| 97 | + |
| 98 | +## Privacy-Max / Local AI Build — First Tier I'd Call "Real Local AI" |
| 99 | + |
| 100 | +**Best for:** Strong privacy posture, local transcription, local agent workflows, and practical small-to-mid local model work. |
| 101 | + |
| 102 | +### Hardware |
| 103 | + |
| 104 | +| Component | Model | Notes | Price | |
| 105 | +|-----------|-------|-------|-------| |
| 106 | +| **Primary Host** | CLX SET Gaming Desktop (Intel Core Ultra 9 285K, 64GB DDR5, GeForce RTX 5070, 2TB NVMe, 4TB HDD) | Heavy compute for local models + agents, excellent cooling | $3,049.99 | |
| 107 | +| **NAS / Online Backup** | Synology DS923+ (4-bay NAS) | 10GbE, NVMe cache, always-on target | $599.99 | |
| 108 | +| **NAS Drives** | 4 × WD Red Plus 8TB (B&H 4-pack) | Full 4-bay RAID, near `50TB` usable | $799.96 | |
| 109 | +| **UPS** | APC BX1500M (1500VA / 900W) | Protects NAS + primary host | $189.99 | |
| 110 | +| **Off-Box Rotation Backup** | WD Elements Desktop 8TB (external HDD) | Rotate off-site | $209.99 | |
| 111 | +| | | **Estimated Total** | **$4,849.92** | |
| 112 | + |
| 113 | +### Why This Is the Local AI Tier |
| 114 | + |
| 115 | +Ollama's published hardware support: |
| 116 | + |
| 117 | +- NVIDIA GPUs: compute capability 5.0+ supported, **RTX 50xx family included** |
| 118 | +- Model memory requirements: |
| 119 | + - **7B models:** ≥8GB VRAM |
| 120 | + - **13B models:** ≥16GB VRAM |
| 121 | + - **70B models:** ≥64GB system RAM |
| 122 | + |
| 123 | +With **64GB RAM + RTX 5070**, this is **the first tier where local AI becomes practical instead of aspirational**: |
| 124 | + |
| 125 | +- Run 7B–13B models comfortably on GPU VRAM |
| 126 | +- Fall back to CPU for 70B models or batched inference |
| 127 | +- Orchestrate multi-model agent workflows locally |
| 128 | +- Private voice, vision, and text processing |
| 129 | + |
| 130 | +### Important Limitation |
| 131 | + |
| 132 | +This is a **small-to-mid local-model tier, not a "run all models locally all the time" tier**. If you specifically want big models running concurrently, move into heavier workstation/server budgets (3× cost and up). |
| 133 | + |
| 134 | +### Typical Deployment |
| 135 | + |
| 136 | +- CLX desktop runs LifeOS core + local Ollama |
| 137 | +- Load 7B–13B models (Llama 2, Mistral, etc.) on the RTX 5070 |
| 138 | +- Use DS923+ for synchronized backups and media library |
| 139 | +- Rotate external WD Elements off-site monthly |
| 140 | +- Local Whisper, vision, and reasoning agents stay private |
| 141 | +- Fall back to cloud only for very large models or exceptional load |
| 142 | + |
| 143 | +--- |
| 144 | + |
| 145 | +## Which One to Choose |
| 146 | + |
| 147 | +### My Recommendation |
| 148 | + |
| 149 | +**Buy the Recommended build unless you already know that local AI is a core requirement on day one.** |
| 150 | + |
| 151 | +- **Budget build** is enough to get LifeOS into real daily service, but it feels scrappy for always-on production. |
| 152 | +- **Recommended build** is the one that feels like a **proper home production node** with sane storage and backup posture. |
| 153 | +- **Privacy-max build** is the right jump **only when you truly want local inference and privacy-first AI to be part of the product, not just an experiment**. |
| 154 | + |
| 155 | +### Cost Reduction Order |
| 156 | + |
| 157 | +If you need to cut cost **within** a tier, cut in this order: |
| 158 | + |
| 159 | +1. **Skip the NAS on the Budget build** |
| 160 | +2. **Keep the UPS** (non-negotiable) |
| 161 | +3. **Keep at least one separate backup target** (non-negotiable) |
| 162 | +4. **Do not overbuy local-AI hardware until local models are truly part of your plan** |
| 163 | + |
| 164 | +This keeps the system **professional and resilient** without forcing you into the most expensive tier too early. |
| 165 | + |
| 166 | +--- |
| 167 | + |
| 168 | +## Key Assumptions |
| 169 | + |
| 170 | +- **Expo EAS Build:** Mobile app binaries (iOS/Android) are built in Expo's hosted service. You do not need a Mac or iOS device for CI/CD. |
| 171 | +- **Always-on expectation:** Target 99% uptime for personal use (UPS + single backup disk is acceptable; NAS + RAID adds the second layer). |
| 172 | +- **Self-hosted:** You own the hardware, data, and encryption keys. Cloud AI is available but optional. |
| 173 | +- **Scalability:** Each tier supports growth to the next one without total rebuild: |
| 174 | + - Budget → Recommended: Add NAS, keep the Beelink |
| 175 | + - Recommended → Privacy-max: Upgrade primary host, keep NAS and backup discipline |
| 176 | + |
| 177 | +--- |
| 178 | + |
| 179 | +## Related Documentation |
| 180 | + |
| 181 | +- [Hardware Profile](./hardware.md) — Reference architecture and design philosophy |
| 182 | +- [SETUP.md](../SETUP.md) — Development environment setup |
| 183 | +- [Security and Privacy](./security-and-privacy.md) — Encryption, backup, and data protection strategy |
0 commit comments