# HOME — Local Agent OS (Tony's Windows PC)

> **Canonical vault:** `~/Documents/Obsidian Vault` (this folder)
> **Agent:** Hermes (local PC profile) + Leonard personality
> **Date started (local setup):** 2026-08-28
> **Current primary models:** OpenRouter DeepSeek (deepseek-v4-flash-0731, current), FreeToken local (Gemma-4-26B + gpt-oss-20b), remote brother boxes via Tailscale, local llama-cpp at 1920

---

## North Star

One Obsidian vault that:
- You browse and edit directly on the laptop
- Hermes reads from and writes to for compounding memory
- Survives model changes (local FreeToken ↔ Grok ↔ remote)
- Uses the Personal Agent OS model from `Tony_tech/personal-agent-os`

This is the external brain: raw input → structured notes → project status → agent lessons.

## Current Structure (Agent OS + PARA hybrid)

| Path | Purpose |
|------|---------|
| `index.md` | Vault index — link map to everything |
| `HOME.md` | Constitution — north star, active projects, non-negotiables |
| `STATUS.md` | Health snapshot — services, cron jobs, model stack |
| `00_Raw/` | Dump zone — transcripts, quick captures, voice blobs |
| `00_system/` | Memory policy, boot instructions, decision log |
| `01_user/` | Durable facts about you (confirmed + proposed) |
| `01_Wiki/` | High-signal curated topic pages |
| `02_agents/` | Hermes + Leonard rules, lessons, boot prompts |
| `03_shared/` | Cross-agent workflows (ready for use) |
| `04_Daily/` | Daily notes & consolidations |
| `04_projects/` | Active projects (status.md + decisions.md per project) |
| `05_Output/` | Shippable artifacts for humans/clients |
| `09_templates/` | Note templates (project, KB entry, experiment log) |
| `10_archive/` | Stale/old projects |
| `10_Projects/` | Personal overflow (PARA) |
| `20_Areas/` | Personal areas (PARA) |
| `30_Resources/` | Personal resources (PARA) |

## Active Projects

See `04_projects/` for status.

Key current:
- **VeriPath** — human-supervised business verification / audit service (core business)
- **Athena Oracle** — AI research pipeline (GitHub + arXiv + HF ingestion, daily digests)
- **Hermes Agent local PC setup** — complete (model stack, vault RAG, auto-start, watchdog, TTS, desktop, kanban)
- **Thornhill Construction website** — built, deployment pending DNS decision
- **Local inference stack** — LFM2.5 primary, FreeToken manual, brother nodes via tailnet

## Non-negotiables

- Secrets, tokens, keys stay in `~/Documents/Vault` (never in vault)
- Only durable, high-signal information goes into project/wiki/agent memory
- User memory: propose in `01_user/proposed_memory.md`, wait for confirmation
- Prefer LFM2.5 (local, fast, agentic+reasoning) for daily work; use brother-5090 for complex reasoning that exceeds 65K context
- Vault RAG uses LM Studio embeddings at :1234 + ChromaDB; reindex after any new notes are added
- FreeToken requires manual start; cannot coexist with LFM2.5 on 12GB GPU (needs ~15GB combined)
- Git the important parts or sync via the Gitea mirror when needed
- All agents follow `02_agents/WRITE_PROTOCOL.md` (write high-signal memory; do not rely on the user to maintain notes).

## How Hermes Uses This Vault (Memory Layer)

Hermes (via obsidian skill + file tools):
- Loads `HOME.md` + `00_system/memory_policy.md` + relevant `04_projects/*/status.md` + `02_agents/leonard/lessons.md` at start of complex sessions
- Searches with `search_files` (target files or content)
- Writes status updates, decisions, daily notes, lessons
- Uses `write_file` / `patch` for precise edits

**Boot order for multi-step work (keep under ~1500 tokens unless needed):**
1. This HOME.md
2. `02_agents/WRITE_PROTOCOL.md`
3. `00_system/memory_policy.md`
4. `02_agents/leonard/lessons.md` (or relevant agent lessons)
5. Relevant project `status.md` or `decisions.md`

## Model & Inference Reality (Local PC)

- **Primary:** LFM2.5-2.6B via llama.cpp at `:1920` (65K ctx, ~145 tok/s, agentic+reasoning)
- **Fallback:** brother-5090 Qwen3.6 27B MTP via tailnet `100.64.0.2:8084` (200K ctx, 27B params)
- **Secondary local:** FreeToken gpt-oss-20b at `:1919` (manual start, can't coexist with LFM2.5 on 12GB GPU)
- **Cloud backup:** OpenRouter free models (Gemma-4-26B)
- **Voice & TTS:** Wake word "hey Jimmy" (sherpa), Edge TTS AriaNeural, auto_tts=true, barge-in enabled, ffplay v9.0.1
- Auto-start: `start-models.bat` in Windows Startup (LFM2.5 launches on boot)
- Watchdog: `lfm25-watchdog` cron (every 2m) auto-relaunches LFM2.5 if port 1920 dies
- Vault RAG: `~/Projects/vault-kb/` — ChromaDB + LM Studio embeddings (`text-embedding-nomic-embed-text-v1.5` at :1234), 240 chunks / 43 files indexed
- Git sync: `vault-sync` cron (every 6h) pushes vault to Gitea private mirror

## Getting Started / Daily Loop

1. Dump to `00_Raw/` or Inbox
2. Consolidate into project `04_projects/<slug>/status.md`
3. Record decisions in `decisions.md`
4. Hermes writes lessons to `02_agents/leonard/lessons.md`
5. Daily work → `04_Daily/`
6. Review `HOME.md` and project status before big sessions

See the Gitea mirror at `~/Documents/AgentOS/` for full templates and older knowledge-base.

## Project Creation Template

Inside `04_projects/<kebab-name>/`:
- `status.md` — current state, goals, next actions
- `decisions.md` — dated decisions with rationale
- Any supporting notes, evidence, specs

Use wikilinks `[[HOME]]`, `[[VeriPath]]` liberally.

---

This vault is now the primary long-term memory for your local Hermes Agent. Load it. Use it. Let it compound.