# HOME — Tony Second Brain / AgentOS

> **Canonical vault root:** `/home/vpsadmin/AgentOS`  
> **Open this folder in desktop Obsidian.**  
> Last updated: 2026-07-09  
> Maintainer: Anthony (Tony) · Operator agent: Leonard (Hermes profile)

---

## North star

One local markdown vault that:

1. You can browse in Obsidian on the laptop  
2. Leonard can read/write on the VPS (no root required)  
3. Survives model swaps (system > model)  
4. Compounds: Raw dump → structured notes → project status

This is the Julian-Goldie-shaped loop, Hermes-native: **vault = memory galaxy**.

---

## Active projects (max 3 primary)

| Project | Path | Status (see status.md) |
|---------|------|------------------------|
| **Athena Oracle** | `04_projects/athena-oracle/` | Live research pipeline |
| **AI Talk Show** | `04_projects/ai-talk-show/` | Dual agents + edge-tts; native TTS on laptop if desired |
| **Local velocity / Phoenix** | `04_projects/local-velocity/` + `phoenix-salon/` | Salon digital presence |

Secondary / parked: memory-system tests (complete), Grok inbox experiments (do not expand).

---

## Non-negotiables

- **No root required** for vault work — everything under `$HOME`  
- **Gitea:** `Tony_tech/*` and `Leonard/*` writable; `Ty_Tech/*` read-only  
- **No firehose** — selective saves into the right project folder  
- **User memory:** propose only → `01_user/proposed_memory.md` (Tony confirms)  
- **Secrets:** never in the vault (tokens stay in env / agent-config outside git if sensitive)  
- **Grok:** not for agent/MoA calls (OOC); free/local models only for tooling  
- **VPS Chatterbox / swap:** off agenda — laptop optional for local TTS  

---

## How the loop works

```
YOU (laptop Obsidian)     LEONARD (VPS Hermes)      GITEA (optional mirror)
        │                        │                         │
   edit notes ◄── Git/Syncthing ──► read/write vault  ──push──► personal-agent-os
   dump to 00_Raw/          consolidate after work           backup/history
```

1. Dump messy stuff → `00_Raw/`  
2. Durable project truth → `04_projects/<name>/status.md` + `decisions.md`  
3. Topic synthesis → `01_Wiki/` (one idea / topic page when worth it)  
4. Daily briefs / session notes → `04_Daily/`  
5. Shippable artifacts → `05_Output/`  
6. Agent lessons → `02_agents/leonard/lessons.md` (reusable only)

---

## Agent boot (Leonard)

On multi-step sessions, load in order (token budget ~1500 unless task needs more):

1. This file (`HOME.md`) — skim  
2. `00_system/memory_policy.md`  
3. `02_agents/leonard/lessons.md`  
4. Relevant `04_projects/<project>/status.md`

Full checklist: `02_agents/leonard/boot_prompt.md`

---

## MoA policy

Default: **Leonard builds alone.**  
Escalate to mixture-of-agents only when Leonard flags a real design fork (e.g. sync architecture failure modes, product vs service tradeoffs). Not for folder creation or routine consolidations.

---

## Desktop setup

See `DESKTOP_SYNC.md`.