AI Systems
One interface. Five systems. Shared context.
AgentHub is the front door to the systems I use every day. Obsidian keeps their knowledge persistent and shared.
AgentHub
Unified human ↔ agent interface
Five peer systems. Independent, and able to exchange information.
-
PersonalOS
Goals, finance, routines, and life admin.
-
CareerOS
Learning, interviews, roles, and follow-through.
-
FantasyOps
Data, analysis, and recommendations.
-
Homelab Ops
Services, health, automation, and incidents.
-
Local AI Platform
Routing, inference, and local + cloud compute.
Obsidian
Shared Knowledge Base
persistent context • documentation • decisions • plans • runbooks • learnings • agent handoffs
Independent systems • shared memory • durable interfaces
How it works
From request to remembered result.
-
01
Request work through AgentHub
-
02
Route it to the appropriate agent, system, or compute
-
03
Use domain workflows and shared Obsidian context
-
04
Return the result and persist useful knowledge
Local AI Platform
Local and cloud intelligence, routed per task.
Local
Routine, privacy-sensitive, and available workloads.
Cloud
Specialized, high-complexity workloads.
Routed by
- Capability
- Availability
- Latency
- Cost
- Privacy
Operating footprint
The scale it runs on.
- 152 GB
- unified memory across two local AI hosts
- 6.5 TB
- local storage capacity across four tiers
- ≤70B DENSE*
- primary local LLM workloads
- 4 tiers
- AI inference, system/control, shared knowledge, photography
- GB10 128 GB unified memory / 2 TB NVMe Primary AI compute, ≤70B dense
- Apple Silicon 24 GB unified memory / 512 GB SSD Control plane + lightweight inference, under 20B
- HomelabHD 2 TB Shared vault, development, homelab data
- PhotographySSD 2 TB Originals, exports, photo archive
Available on demandAMD GPU workstation · burst inference · benchmarking · runtime experimentation
Capacity, not utilization. Memory is physical per host, not pooled.
* MoE models can exceed 70B total parameters when their quantized weights fit local memory; only a subset of experts is active per token.
