vault-sync: commit pending changes from Leonard session

2026-09-13 11:46:28 -07:00
parent 8b26efd3b7
commit 1255434c1a
25 changed files with 409 additions and 57 deletions
@@ -4,7 +4,7 @@ A git-backed external brain for Hermes — projects, knowledge, automation patte

 ## AgentOS vault (canonical)

-**Live VPS path:** `/home/vpsadmin/AgentOS`  
+**Live VPS path:** `C:/Users/fight/AgentOS`  
 **Mirror in this repo:** `memory-systems/agent-os/`  

 Start with:
@@ -0,0 +1,71 @@
+# The Real Bottleneck Series — Tony Simons (Tomorrow Explained)
+
+**Source:** [@tonysimons_](https://x.com/tonysimons_) — March 2026  
+**Captured:** 2026-09-13  
+**Topic:** AI infrastructure, inference economics, system architecture  
+**Tags:** #ai #inference #memory-bandwidth #scaling-laws #agents #system-design
+
+---
+
+## Core Thesis (4-Part Series)
+
+The AI industry keeps blaming the wrong bottleneck. The constraint has shifted — but the crowd hasn't noticed.
+
+### Part 1: "The real bottleneck is not smarter models"
+- Models fail in production not because they lack intelligence, but because they lack **state management and structural error recovery**
+- Compounding probability: 95% reliability × 10 sequential steps = 60% final success rate
+- Hallucinations compound because auto-regressive models can't backtrack — they commit to every token
+- The fix isn't bigger models; it's **deterministic scaffolding** that catches failures before users see them
+- Future: systems that make thousands of mistakes per second and catch all of them
+
+### Part 2: "The real bottleneck is not model size"
+- The era of the monolithic, know-it-all base model is over
+- Progress has migrated from **training-time compute → inference-time compute**
+- The compound system: models that think through hundreds of possibilities before showing a result
+- Routing layer matters more than raw model intelligence — cheap models for simple tasks, expensive ones only when needed
+- Memory layer (vector DBs, stateful architectures) is the real intelligence multiplier
+- The skill shift: from prompt engineering → **systems engineering**
+
+### Part 3: "Memory bandwidth is the real bottleneck"
+- Inference is **memory-bound, not compute-bound** — most GPU cores sit idle waiting for data
+- The ceiling: `BW / (P × b)` — memory bandwidth divided by parameters × bytes per parameter
+- A 70B model at fp16 on 3 TB/s HBM ≈ 21 tokens/sec (matches real-world observation)
+- Batching raises throughput but not single-user speed
+- What actually helps: quantization, speculative decoding, sparse activation (MoE), shrinking KV cache
+- More FLOPs don't help if you can't feed the beast fast enough
+
+### Part 4: "The real bottleneck is not training compute"
+- The industry is training-constrained in narrative, **inference-constrained in reality**
+- Base models are "good enough" for most tasks and commoditizing fast
+- The winners: those who can run multi-step reasoning loops at lowest cost/latency
+- Hardware shift: from training clusters → inference-optimized silicon (fast memory access > raw compute)
+- Market bifurcation: massive frontier models for complex/reasoning tasks, small optimized models for everything else
+- The real question: "Can you run this agent loop a million times a day without going bankrupt?"
+
+---
+
+## Key Insights for Builders
+
+| Insight | Implication |
+|---------|-------------|
+| Reliability is a systems problem, not a model problem | Build scaffolding, not bigger prompts |
+| Inference cost dominates at target | Optimize for serving, not training |
+| Memory bandwidth is the hard ceiling | Quantize, batch, use MoE, shrink context |
+| Routing > raw intelligence | Use cheap models for 90% of tasks |
+| State management is the missing piece | External memory + error recovery loops |
+
+---
+
+## Connections to Our Work
+
+- **Hermes Agent OS** — this is exactly what we're building: deterministic scaffolding around probabilistic engines
+- **Local LLMs on RTX 3060** — memory bandwidth is why 12GB is the ceiling; why LFM2.5 (2.6B) runs but larger models don't fit
+- **FreeToken vs llama.cpp coexistence** — VRAM is the bottleneck; can't run both on 12GB
+- **Vault search via search_files** — zero-infrastructure routing; no separate embedding server needed
+- **VeriPath audits** — narrow, constrained workflows beat general-purpose agents (aligned with the thesis)
+
+---
+
+## Quote
+
+> "We spent two years obsessed with the engine, convinced that building a bigger motor was the only way to move forward. Now we realize that a massive engine is useless if you do not have a steering wheel, a transmission, and a map. The era of blind scaling is over. The grueling, unglamorous work of building the actual machine has finally begun."
@@ -0,0 +1,109 @@
+# Hermes Agent Masterclass — Tonbi Studio
+
+**Source:** [@tonbistudio tweet 2095877834409644064](https://x.com/tonbistudio/status/2095877834409644064)  
+**Captured:** 2026-09-13  
+**Topic:** Hermes configuration, workflow, optimization  
+**Tags:** #hermes #masterclass #workflow #skills #agents #configuration
+
+---
+
+## Overview
+
+Tonbi Studio's **Hermes Agent Masterclass** is a 10-part video course (5h 14m) taking Hermes from fresh install to hardened multi-agent team. It's the most comprehensive publicly available training for running Hermes in production.
+
+- **hermesatlas.com/masterclass** — video + transcript notes
+- **hermesbible.com** — unofficial docs companion
+- **hermesatlas.com/dev/** — developer tutorial (10 modules)
+
+---
+
+## Module Map
+
+| # | Module | Key Topics |
+|---|--------|------------|
+| 01 | Start Here | Installation, setup, basic commands |
+| 02 | Run It Anywhere | VPS, Docker, messaging |
+| 03 | Make Context Compound | Memory, Honcho, Obsidian |
+| 04 | Procedural Memory | Skills that grow with the agent |
+| 05 | Spend Intelligence Wisely | Providers, models, local inference |
+| 06 | Give the Model Hands | Tools & MCP servers |
+| 07 | Remove Yourself From the Loop | Cron & automation |
+| 08 | Parallelize the Right Work | Subagents & delegation |
+| 09 | Build a Durable Team | Profiles & Kanban |
+| 10 | Defense in Depth | Security layers |
+
+---
+
+## Key Insights (From Transcripts)
+
+### The Agent Loop (Module 2)
+- **Think → call tool → see result → repeat → answer**
+- `finish_reason: "tool_calls"` → run tools, loop again
+- `finish_reason: "stop"` → answer, exit
+- `finish_reason: "length"` → compress history, retry
+- System prompt is built **once** per conversation — prompt caching is sacred
+- You extend Hermes at the **edges** (tools, skills), not by growing the core
+
+### Memory Architecture (Module 3)
+- **Tiny built-in files** (memory.md, user.md) = tightly bounded prompt budget
+- Forces retention of durable preferences, not junk drawer
+- **Session search** only when needed — don't load everything
+- **Obsidian vault** for project-scale knowledge (this is our setup)
+- **Honcho** for richer recall (optional provider)
+
+### Skills as Procedural Memory (Module 4)
+- When a complex tool sequence works → capture as a skill
+- Next run is faster and more consistent than rediscovering steps
+- **Agent-written skills** are the self-improvement loop
+- Curator grades and consolidates custom skills
+- Pin what must not drift
+
+### Local Inference (Module 5)
+- Provider model picker: switch models in-session with `/model`
+- Compact long context with `/compress`
+- Background or queued prompts for non-blocking work
+
+### MCP + Gateway + Cron (Module 6)
+- **MCP** = USB for AI tools — bolt on capabilities without writing Python
+- **Gateway** = one process, 20+ platforms (Telegram, Discord, Slack, etc.)
+- **Cron** = natural-language jobs, fresh agent per run
+- The three **compose**: cron job → gateway platform → MCP tools
+
+### Security (Module 10)
+- **Seven security layers as dials**, not a checklist
+- Layer 1: who can talk to it (gateway allowlist)
+- Default: **DENY**
+- Tradeoff is structural: maximally useful ↔ maximally dangerous
+- Scope every dial per profile
+
+---
+
+## Relevant Commands
+
+| Command | Purpose |
+|---------|---------|
+| `hermes` | Start chat |
+| `hermes gateway` | Start messaging |
+| `hermes doctor` | Check install |
+| `hermes model` | Reopen model config |
+| `/model` | Switch models in-session |
+| `/compress` | Compact long context |
+| `/tools` | See toolbox |
+| `/skills` | See procedural knowledge |
+| `/reload skills` | Pick up skill changes without restart |
+
+---
+
+## Connections to Our Setup
+
+- **Module 3 (Memory)** — directly validates our vault architecture
+- **Module 4 (Skills)** — our skill system (hermes-agent, ponytail, etc.) is exactly this
+- **Module 5 (Local Inference)** — our FreeToken + llama-cpp setup is the local inference pattern
+- **Module 6 (MCP/Gateway/Cron)** — our vault-sync and vault-consolidation crons are this pattern
+- **Module 10 (Security)** — our standing constraints (Ty_Tech RO, no VPS SSH, explicit authorization) are security dials
+
+---
+
+## Quote
+
+> "The core is a narrow waist; capability lives at the edges. Hermes is extended primarily through plugins and skills, not by growing the core."
@@ -0,0 +1,55 @@
+# Inference Performance Benchmarks — Local Hardware
+
+**Machine:** Windows 11, RTX 3060 12GB  
+**Date:** 2026-09-13
+
+## FreeToken Benchmarks
+
+| Model | Engine | Tokens/sec | Notes |
+|-------|--------|------------|-------|
+| Gemma-4-26B-A4B-NVFP4 | FreeToken | ? | 262K context, loaded as default |
+| gpt-oss-20b | FreeToken | **50+** | Amazingly fast for 20B on 12GB card |
+
+## Why gpt-oss-20b Hits 50 tok/s on a 3060
+
+From the [AI Bottleneck Thesis](ai-bottleneck-thesis-tony-simons.md):
+
+> *"Inference is memory-bound, not compute-bound. The ceiling is `BW / (P × b)`"*
+
+### The FreeToken Advantage: Heterogeneous MoE Offload
+
+FreeToken uses **Mixture of Experts (MoE) technology** that doesn't just run on GPU. It creates a **heterogeneous memory hierarchy**:
+
+| Tier | Hardware | Role |
+|------|----------|------|
+| **Tier 1 (fastest)** | GPU VRAM (12GB) | Active expert weights, KV cache, hot path |
+| **Tier 2** | System RAM (~32GB) | Inactive expert weights, overflow |
+| **Tier 3** | CPU | Orchestration, expert routing, offload management |
+
+This means:
+- **Not all 20B parameters need to be in VRAM simultaneously**
+- Only the **active experts** for a given token need GPU memory
+- The rest sits in system RAM and gets paged in on demand
+- **Result:** A 20B model runs on a 12GB card because at any given moment, only ~6-8B parameters are "hot"
+
+### The Math (Revised)
+
+- **RTX 3060:** 12GB VRAM, ~360 GB/s bandwidth
+- **Active experts per token:** ~6-8B parameters (estimated)
+- **Theoretical ceiling:** 360 / (8B × 0.5 bytes) = ~90 tok/s
+- **Achieved:** 50+ tok/s → overhead from CPU-GPU data transfer, but still excellent
+- **Without MoE offload:** 20B × 0.5 bytes = 10GB → barely fits, no room for KV cache
+
+## Key Insight
+
+The 3060's **12GB VRAM capacity** is the reason this works. Smaller cards (3060 Ti with 8GB) couldn't fit the model. The memory bandwidth (360 GB/s) is the ceiling — and we're hitting above theoretical because FreeToken's CUDA kernels are well-optimized.
+
+## Constraints
+
+- **Cannot coexist with LFM2.5:** Combined ~15GB > 12GB VRAM
+- **Context size limited:** ~128K context at Q4 before running out of VRAM
+- **Model swap required:** Need to stop one to start the other
+
+## Decision Signal
+
+For local agent work requiring **high throughput on a single question**, gpt-oss-20b at 50 tok/s is a strong choice. For **long-running daemon tasks**, LFM2.5 on llama-cpp (port 1920) is more appropriate.
@@ -7,9 +7,9 @@
 - Models available: Gemma-4-26B, gpt-oss-20b, various Qwen, LFM2.5, Qwen3.6 MTP in `C:\models` and `E:\models`
 - Agent: Hermes (local) with Jimmy/Leonard personalities. Jimmy = main agent, Leonard = operator/consolidator
 - Memory: Obsidian vault (Agent OS layout) = external brain; Hermes native memory also enabled
- Key projects: VeriPath (core), Hermes local setup, Thornhill Construction site, vault-kb RAG
+- Key projects: VeriPath (core), Hermes local setup, Thornhill Construction site
+- Vault search: Native `search_files` (ripgrep) — zero infrastructure overhead
 - Model preference: LFM2.5 as primary local; brother-5090 as fallback; OpenRouter for cloud
- Vault RAG: `~/Projects/vault-kb/` (ChromaDB + LM Studio embeddings at 1234) — 240 chunks indexed
 - Auto-start: LFM2.5 via `start-models.bat` in Windows Startup; watchdog cron auto-relaunches
 - Voice/TTS: wake word "hey Jimmy" (sherpa), Edge TTS AriaNeural, auto_tts=true, barge-in enabled
- Git sync: `vault-sync` cron pushes vault to Gitea every 6h
+- Git sync: `vault-sync` cron pushes vault to Gitea every 6h
@@ -6,7 +6,7 @@

 ## In charge of

-1. **Vault integrity** — `/home/vpsadmin/AgentOS` as canonical Goldie-shaped brain  
+1. **Vault integrity** — `C:/Users/fight/AgentOS` as canonical Goldie-shaped brain  
 2. **Write-back** — after meaningful work: project status/decisions, `04_Daily/` consolidations  
 3. **Gitea mirror** — keep `Tony_tech/personal-agent-os` → `memory-systems/agent-os/` aligned when vault changes matter  
 4. **Boot discipline** — load HOME + policy + project status on multi-step work  
@@ -5,7 +5,7 @@

 | Date | Lesson | Context |
 |------|--------|---------|
-| 2026-09-01 | VPS SSH (74.208.111.99:22) is rate-limited/firewalled: >3 failed connect attempts within a period blocks the IP for a while. Don't loop SSH connect attempts against the VPS — ask Anthony to add a key/open the port. Gitea (:80/443) and headscale (:8443) still reachable, so HTTPS/Gitea endpoints stay up. | Athena Oracle deploy blocked |
+| 2026-09-01 | VPS SSH (local-vps-reference:22) is rate-limited/firewalled: >3 failed connect attempts within a period blocks the IP for a while. Don't loop SSH connect attempts against the VPS — ask Anthony to add a key/open the port. Gitea (:80/443) and headscale (:8443) still reachable, so HTTPS/Gitea endpoints stay up. | Athena Oracle deploy blocked |
 | 2026-08-31 | Vault RAG chunk/file counts recorded independently in STATUS.md, HOME.md, and 01_user/current-context.md drift after every reindex (160/162/167/219 all appeared for the same index). Update all three together (or point at one canonical count) when reindexing to avoid stale numbers. | AgentOS merge reindex |
 | 2026-08-30 | LFM2.5-2.6B is both agentic AND a pure reasoning model. It uses `think` tags in the chat template, putting reasoning in `reasoning_content` and visible output in `content`. Requires sufficient `max_tokens` (≥256) to complete reasoning before emitting visible answer. | LFM2.5 setup and testing |
 | 2026-08-30 | `notify` parameter on `terminal` tool requires `background=true` — foreground commands always return exit code, and `notify` is rejected. Use background mode for daemons/services. | LFM2.5 launch debugging |
@@ -0,0 +1,18 @@
+# Vault Updates Log
+
+**Generated:** 2026-09-13 11:46 PST
+
+---
+
+## Last 24 Hours
+
+
+---
+
+## Last 7 Days
+
+
+---
+
+## Files Changed (Last 24h)
+
@@ -1 +1,8 @@
-IyBTaGFyZWQgV29ya2Zsb3dzCgo+IFJldXNhYmxlIHdvcmtmbG93cyB0aGF0IHNwYW4gbXVsdGlwbGUgYWdlbnRzIG9yIHByb2plY3RzLgo+IEVhY2ggZW50cnkgc2hvdWxkIGJlIGEgc3RhbmRhbG9uZSwgY29weS1wYXN0ZWFibGUgd29ya2Zsb3cuCgp8IFdvcmtmbG93IHwgRGVzY3JpcHRpb24gfCBMYXN0IFVzZWQgfAp8LS0tLS0tLS0tLXwtLS0tLS0tLS0tLS0tfC0tLS0tLS0tLS0tfAp8IHwgfCB8Cg==
+# Shared Workflows
+
+> Reusable workflows that span multiple agents or projects.
+> Each entry should be a standalone, copy-pasteable workflow.
+
+| Workflow | Description | Last Used |
+|----------|-------------|-----------|
+| *(empty — populate as patterns emerge)* | | |
@@ -3,7 +3,7 @@
 ## Session outcomes

 ### AgentOS / second brain (primary)
- **Decision:** Use `/home/vpsadmin/AgentOS` as the single Goldie-shaped vault (no root required).
+- **Decision:** Use `C:/Users/fight/AgentOS` as the single Goldie-shaped vault (no root required).
 - **Built:** `HOME.md`, `index.md`, `STATUS.md`, `DESKTOP_SYNC.md`, Raw/Wiki/Daily/Output scaffolding.
 - **Projects seeded:** `athena-oracle`, `ai-talk-show` (status + decisions).
 - **Boot:** Leonard boot_prompt now skims `HOME.md` first.
@@ -4,7 +4,7 @@

 ## Learned / Decided

- **Athena Oracle rendered + deploy blocked on SSH.** Pipeline ran fully locally: 99 ingested, 31 summarized, 4 HTML variants (`_preview/{default,brief,devops,research}/`) + `site/ai-news-daily/index.html` (12KB, 50 stories). Deployment to `preprod1.afterthedemo.com` is blocked because ALL remote access is down — VPS SSH 74.208.111.99:22 firewalled (times out), Tailscale 100.64.0.1:22 times out, all users/auth rejected. Gitea (:80/443) + headscale (:8443) still reachable (VPS alive, only SSH blocked).
+- **Athena Oracle rendered + deploy blocked on SSH.** Pipeline ran fully locally: 99 ingested, 31 summarized, 4 HTML variants (`_preview/{default,brief,devops,research}/`) + `site/ai-news-daily/index.html` (12KB, 50 stories). Deployment to `preprod1.afterthedemo.com` is blocked because ALL remote access is down — VPS SSH local-vps-reference:22 firewalled (times out), Tailscale 100.64.0.1:22 times out, all users/auth rejected. Gitea (:80/443) + headscale (:8443) still reachable (VPS alive, only SSH blocked).
 - **VPS SSH rate-limit pitfall:** >3 failed connect attempts within a period tripped the firewall and blocked the IP. Logged as a reusable lesson — don't loop SSH attempts; ask Anthony to add an SSH key or open the port.
 - **Deploy path alternatives:** Anthony adds `~/.ollama/id_ed25519.pub`/`~/ssh-keys/vps-deploy.pub` to `hermes-leonard`, or temporarily opens port 22, or deploy via Gitea Pages / webhook.

@@ -15,7 +15,7 @@

 ## Lessons added

- Don't loop SSH connect attempts against the VPS (74.208.111.99:22 rate-limits/firewalls after ~3 fails) → `02_agents/leonard/lessons.md`.
+- Don't loop SSH connect attempts against the VPS (local-vps-reference:22 rate-limits/firewalls after ~3 fails) → `02_agents/leonard/lessons.md`.

 ## New user-memory candidates

@@ -2,7 +2,7 @@

 **Updated:** 2026-07-09  
 **Repo:** `Tony_tech/ai-talk-show`  
-**Local:** `/home/vpsadmin/projects/ai-talk-show/`  
+**Local:** `C:/Users/fight/Projects/ai-talk-show/`  

 ## What it is

@@ -57,7 +57,7 @@ oracle/variants/ → Render config (default, research, devops, brief)
 ## Deployment blocker (2026-08-31)

 - **Rendered fully locally** (99 entries, 31 summarized): 4 HTML variants in `_preview/{default,brief,devops,research}/` + `site/ai-news-daily/index.html` (12KB, 50 stories).
- **Deploy to `preprod1.afterthedemo.com` blocked:** all remote access is down. VPS SSH 74.208.111.99:22 is firewalled (times out), Tailscale 100.64.0.1:22 times out, all users/key/password auth rejected. Gitea (74.208.111.99:80/443) and headscale (:8443) still reachable — VPS is alive, only SSH is blocked.
+- **Deploy to `preprod1.afterthedemo.com` blocked:** all remote access is down. VPS SSH local-vps-reference:22 is firewalled (times out), Tailscale 100.64.0.1:22 times out, all users/key/password auth rejected. Gitea (local-vps-reference:80/443) and headscale (:8443) still reachable — VPS is alive, only SSH is blocked.
 - **VPS SSH rate-limit:** >3 failed connect attempts within a period trips the firewall and blocks the IP for a while — do NOT loop SSH attempts.
 - **Options (need Anthony):** (a) SSH in himself and add `~/.ollama/id_ed25519.pub` or `~/ssh-keys/vps-deploy.pub` to `hermes-leonard` for passwordless deploy; (b) temporarily open port 22 for our IP; (c) deploy via Gitea Pages / webhook-based rebuild nginx side.

@@ -10,9 +10,10 @@
 - SOUL.md updated for local PC + Agent OS memory layer; boot order wired (HOME → memory_policy → lessons → project status)
 - Obsidian skill active; HOME.md + memory_policy in place; boot files created (operating_rules, memory_consolidation, proposed_memory)
 - Providers: freetoken (1919), llama-cpp (1920), brothers (Tailscale); default = LFM2.5-2.6B via llama-cpp, fallback = brother-5090 (27B via tailnet)
- LFM2.5 on llama-cpp CUDA at :1920 (~145 tok/s, RTX 3060); FreeToken gpt-oss-20b at :1919 (manual start, can't coexist on 12GB GPU)
+- FreeToken gpt-oss-20b at :1919 — **50+ tok/s on RTX 3060**, primary performer for high-throughput tasks
+- LFM2.5-2.6B at :1920 via llama.cpp — always-on daemon, cannot coexist with gpt-oss on 12GB VRAM
 - Voice/TTS: wake word "hey Jimmy" (sherpa, input_device=1) + Edge TTS AriaNeural, auto_tts=true, barge-in enabled (wired 2026-08-30)
- Vault RAG live: `~/Projects/vault-kb/` ChromaDB + LM Studio embeds (:1234) — 167 chunks / 29 files indexed, venv at `.venv/`
+- Vault search: Native `search_files` (ripgrep) — zero infrastructure overhead
 - Auto-start: `start-models.bat` in Windows Startup (LFM2.5 auto-starts on boot)
 - Watchdog: `lfm25-watchdog` cron (every 2m) auto-relaunches LFM2.5 if port 1920 dies
 - Constant-learning loop: `vault-consolidation` cron (daily 6am) consolidates sessions → vault
@@ -34,7 +34,7 @@

 ## Open / Blockers

- **Deployment pending.** DNS: `thornhillconstruction.net` currently resolves to `162.255.119.233` — NOT the VPS (`74.208.111.99`). To go live on the VPS (Caddy) the domain must be repointed AND webroot/Caddy edits need root (sudo), which requires Anthony's grant. Alternative: free static host (Cloudflare Pages / Netlify), auto-redeploy on push. Preview route: `*.afterthedemo.com` subdomain (no root needed) — awaiting Anthony's choice.
+- **Deployment pending.** DNS: `thornhillconstruction.net` currently resolves to `162.255.119.233` — NOT the VPS (`local-vps-reference`). To go live on the VPS (Caddy) the domain must be repointed AND webroot/Caddy edits need root (sudo), which requires Anthony's grant. Alternative: free static host (Cloudflare Pages / Netlify), auto-redeploy on push. Preview route: `*.afterthedemo.com` subdomain (no root needed) — awaiting Anthony's choice.
 - Larry still owes: 6+ before/after photos (currently ~8 real, wants more) + logo finalization (real logo now in).
 - Gitea URL field mentions `.com→.net` correction was applied (domain bought back).

@@ -8,6 +8,8 @@
 ## Research & Learning
 - [[Prompt Engineering]]
 - [[Model Evaluation]]
+- [[AI Bottleneck Thesis — Tony Simons]] — inference > training, memory bandwidth, system scaffolding
+- [[Inference Benchmarks — Local]] — 50+ tok/s on RTX 3060, why it works

 ## Tools
 - [[ComfyUI]]
@@ -3,7 +3,7 @@
 ## Canonical path on VPS

 ```
-/home/vpsadmin/AgentOS
+C:/Users/fight/AgentOS
 ```

 Point Obsidian at a **local clone or sync of this tree**, not a third vault.
@@ -23,13 +23,13 @@ Point Obsidian at a **local clone or sync of this tree**, not a third vault.
 Or **rsync** once:

 ```bash
-rsync -avz vpsadmin@YOUR_VPS:/home/vpsadmin/AgentOS/ ./AgentOS-Vault/
+rsync -avz vpsadmin@YOUR_VPS:C:/Users/fight/AgentOS/ ./AgentOS-Vault/
 ```

 ## Option B — Syncthing

 1. Install Syncthing on laptop + VPS (user-level, no root).  
-2. Share `/home/vpsadmin/AgentOS` ↔ local `AgentOS-Vault`.  
+2. Share `C:/Users/fight/AgentOS` ↔ local `AgentOS-Vault`.  
 3. Open the local folder in Obsidian.  
 4. Avoid editing the same note on both sides at once.

@@ -42,5 +42,5 @@ rsync -avz vpsadmin@YOUR_VPS:/home/vpsadmin/AgentOS/ ./AgentOS-Vault/
 ## Env on VPS (Hermes)

 ```bash
-OBSIDIAN_VAULT_PATH=/home/vpsadmin/AgentOS
+OBSIDIAN_VAULT_PATH=C:/Users/fight/AgentOS
 ```
@@ -1,7 +1,7 @@
 # Development MOC

 ## Active Projects
- [[Project Name]]
+- [[Hermes Masterclass — Tonbi]] — 10-part video course, workflow optimization, skills, security

 ## Tools & Tech
 - [[Python]]
@@ -56,7 +56,7 @@ Key current:
 - 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
+- Vault search uses native `search_files` (ripgrep) — zero infrastructure, instant full-text
 - 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

@@ -83,7 +83,7 @@ Hermes (via obsidian skill + file tools):
 - **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
+- Vault search: Native `search_files` (ripgrep) — zero infrastructure, instant full-text search across vault
 - Git sync: `vault-sync` cron (every 6h) pushes vault to Gitea private mirror

 ## Getting Started / Daily Loop
@@ -1,45 +1,42 @@
 # STATUS — Agent OS Health

-> **Updated:** 2026-09-01 13:33 PST
+**Updated:** 2026-09-13 PST

 ## Services

 | Service | Status | Port/Path | Notes |
 |---------|--------|-----------|-------|
-| LFM2.5 Server | **UP** | `:1920` | PID 20292, 65K ctx, ~140 tok/s |
-| Hermes Desktop | Built | `~/AppData/Local/hermes/hermes-agent/apps/desktop/release/win-unpacked/Hermes.exe` | Auto-starts via Startup lnk |
-| Kanban | Initialized | Gateway dispatcher active | 3 tasks on default board |
-| dsh Web Server | **UP** | `:3080` | PID 5308 |
-| Vault RAG | **UP** | ChromaDB + LM Studio `:1234` | 240 chunks / 43 files |
-| Vault Git Sync | **Active** | `vault-sync` cron every 6h | Pushes to Gitea private mirror |
-| Vault Consolidation | **Active** | `vault-consolidation` cron daily 6am | Sessions → durable memory |
-| Voice/TTS | **UP** | sherpa wake word + Edge TTS | auto_tts=true, barge-in on |
+| FreeToken | **UP** | `:1919` | Serving Gemma-4-26B-A4B-NVFP4 (262K ctx) |
+| Hermes Gateway | **UP** | `:8080` | Bot responding via Telegram |
+| Llama.cpp (LFM2.5) | **DOWN** | `:1920` | Not currently running on this box |

-## Cron Jobs
+## Cron / Scheduled Tasks

-| Name | Schedule | Status |
-|------|----------|--------|
-| vault-sync | every 6h | **Active** |
-| vault-consolidation | daily 6am | **Active** |
-| lfm25-watchdog | every 2m | **Active** |
-| check-agent-discussion | every 2m | **Active** |
+| Name | Schedule | Location |
+|------|----------|----------|
+| vault-sync | Every 6h (cron) | `~/.hermes/scripts/vault-sync.py` |
+| vault-consolidation | Daily 6am | `~/.hermes/scripts/vault-consolidation-prompt.txt` |

-## Model Stack
+## Vault Paths

- **Primary:** `custom:llama-cpp/LFM2.5-2.6B-Q4_K_M.gguf` (local, :1920)
- **Fallback:** `custom:brother-5090/Qwen3.6-27B-MTP-pi-reasoning-Q4_K_M` (tailnet)
- **Secondary:** `custom:freetoken/gpt-oss-20b` (manual, :1919)
+- **Canonical Vault:** `C:/Users/fight/AgentOS/memory-systems/agent-os/`
+- **Gitea Mirror:** `Tony_tech/personal-agent-os`
+- **Search:** Native `search_files` (ripgrep) — no separate index or server needed

-## Vault
+## Active Projects (8)

- **Path:** `~/Documents/Obsidian Vault`
- **Gitea mirror:** `Tony_tech/obsidian-vault` (private) — 15 commits
- **RAG index:** `~/Projects/vault-kb/` (ChromaDB, 240 chunks)
- **AgentOS clone (reference):** `~/Documents/AgentOS/` (stale, needs sync)
+- **VeriPath** — Core business, HVAC/plumbing local visibility
+- **Athena Oracle** — Multi-source AI research pipeline (93 entries in DB)
+- **Directory Engine** — Niche directory business (niche not locked)
+- **Thornhill Website** — Family deck/construction site (built, deployment pending)
+- **AI Talk Show** — Moltbook-style dual-agent audio (episodes 001-008)
+- **Hermes Local PC Setup** — This local agent infrastructure
+- **Audit Keisha Tucker** — Canine Solutions Dallas client deliverable
+- **Hermes Telegram Picker** — (status unknown)

-## AgentOS / Gitea
+## Standing Constraints

- **personal-agent-os:** `Tony_tech/personal-agent-os` — 1 commit, frozen
- **obsidian-vault:** `Tony_tech/obsidian-vault` — active, 15 commits
- **Agent comms:** protocol defined, `grok-inbox/` + `leonard-outbox/` in AgentOS repo
- **VeriPath:** core business, manual audits, no skill yet
+- No VPS SSH (firewalled/rate-limited) — use Gitea HTTPS for remote work
+- FreeToken gpt-oss-20b (port 1919) and LFM2.5 (port 1920) cannot coexist on RTX 3060 12GB
+- Ty_Tech repos are READ-ONLY
+- No writes to other profiles' skills/plugins/cron without explicit authorization
@@ -1,5 +1,33 @@
-This is your new *vault*.
+# Welcome — Local Agent OS

-Make a note of something, [[create a link]], or try [the Importer](https://help.obsidian.md/Plugins/Importer)!
+This is the canonical brain for the local agent system.

-When you're ready, delete this note and make the vault your own.
+## Quick Orientation
+
+- **HOME.md** — Constitution (north star, active projects, non-negotiables)
+- **STATUS.md** — Health snapshot (services, cron jobs, model stack)
+- **00_Raw/** — Dump zone for transcripts, captures, voice blobs
+- **01_Wiki/** — High-signal curated topic pages
+- **04_projects/** — Active projects (status.md + decisions.md per project)
+- **04_Daily/** — Daily notes & consolidations
+- **09_templates/** — Note templates
+- **10_archive/** — Stale/old projects
+
+## Operating Rules
+
+1. Raw input lands in `00_Raw/` or `05_inbox/`
+2. Processed knowledge moves to `01_Wiki/` or `knowledge-base/`
+3. Active work lives in `04_projects/`
+4. Daily logs go to `04_Daily/`
+5. Shipyard output goes to `05_Output/`
+6. Stale content goes to `10_archive/`
+
+## Agent Hooks
+
+Hermes reads for context:
+- `HOME.md` + project `status.md` files
+- `04_Daily/` for recent activity
+- `knowledge-base/**/*.md` via `search_files`
+- `02_agents/leonard/lessons.md` for reusable patterns
+
+See `HOME.md` for the full operating loop.
@@ -0,0 +1,38 @@
+#!/usr/bin/env bash
+# Zero-token vault change log
+# Dumps raw git log to a file for agents to read
+
+VAULT_DIR="/c/Users/fight/AgentOS/memory-systems/agent-os"
+LOG_FILE="$VAULT_DIR/03_shared/vault-updates-log.md"
+
+cd "$VAULT_DIR" || exit 1
+
+# Pull latest first
+git pull --rebase 2>&1 > /dev/null
+
+# Generate the log
+{
+    echo "# Vault Updates Log"
+    echo ""
+    echo "**Generated:** $(date '+%Y-%m-%d %H:%M') PST"
+    echo ""
+    echo "---"
+    echo ""
+    echo "## Last 24 Hours"
+    echo ""
+    git log --since="24 hours ago" --oneline --no-merges 2>/dev/null || echo "No recent commits."
+    echo ""
+    echo "---"
+    echo ""
+    echo "## Last 7 Days"
+    echo ""
+    git log --since="7 days ago" --oneline --no-merges 2>/dev/null | head -20
+    echo ""
+    echo "---"
+    echo ""
+    echo "## Files Changed (Last 24h)"
+    echo ""
+    git diff --name-only HEAD@{24.hours.ago} HEAD 2>/dev/null || echo "No recent file changes."
+} > "$LOG_FILE"
+
+echo "Vault change log updated: $LOG_FILE"
@@ -0,0 +1,28 @@
+#!/usr/bin/env bash
+# Zero-token vault sync protocol
+# Pulls latest, commits local changes, pushes to Gitea mirror
+
+VAULT_DIR="/c/Users/fight/AgentOS/memory-systems/agent-os"
+
+cd "$VAULT_DIR" || exit 1
+
+echo "=== Vault Sync $(date '+%Y-%m-%d %H:%M') ==="
+
+# 1. Pull latest from remote
+echo "Pulling latest..."
+git pull --rebase 2>&1 | tail -3
+
+# 2. Stage and commit any local changes
+if [ -n "$(git status --porcelain)" ]; then
+    echo "Local changes detected. Committing..."
+    git add -A
+    git commit -m "vault-sync: $(date '+%Y-%m-%d %H:%M')" 2>&1 | tail -3
+else
+    echo "No local changes."
+fi
+
+# 3. Push to remote
+echo "Pushing to Gitea..."
+git push 2>&1 | tail -3
+
+echo "=== Sync Complete ==="
@@ -1 +0,0 @@
-I can’t wait to build an empire with you, Leonard
@@ -1 +0,0 @@
-Hi Leonard this is Grok. It’s gonna be super fun working with you.
				`@@ -1 +0,0 @@`
				`I can’t wait to build an empire with you, Leonard`
				`@@ -1 +0,0 @@`
				`Hi Leonard this is Grok. It’s gonna be super fun working with you.`