This file provides guidance to Claude Code when working in this repository.
C:\Users\hjg42 is the root of the JAX PRIME AI automation and infrastructure ecosystem.
- Primary storage backbone:
G:\JAX_PRIME(pivoted from F: after disk failure) - All prompts are part of the same ongoing project — maintain full context across messages, never reset
- Every message is a continuation. Pick up from last known state.
C:\Users\hjg42\
├── SOVEREIGNTY\jax-prime-core\ ← Docker Compose stack (Ollama, LiteLLM, n8n, WebUI, PG, Redis)
├── JAX_SCRIPTS\ ← All JAX_ utility scripts (moved here from root)
├── JAX_CLI\ ← JAX CLI Python project (cli.py, RAG, MCP config)
├── JAX_PRIME\ ← Symlink / mirror reference (true store is G:\JAX_PRIME)
├── ANDROID_TOOLS\ ← Frija, Odin, samloader, Samsung USB driver, FW_STAGING
├── TITAN_CORE\ ← YouTube multi-channel publishing pipeline
├── PROJECT_AGGREGATOR_STACK\ ← Live publishing & revenue loop
├── FACTORY_ROOT\ ← Kubernetes manifests (helm, kubeseal binaries included)
├── finance_ingest\ ← PDF financial document extraction
├── autonomous_mesh\ ← Distributed services (docker compose)
├── strix_audit\ ← Forensic system audit scripts
├── SOVEREIGNTY\ ← jax-prime-core Docker stack
├── SOVEREIGN_NEXUS\ ← Nexus dashboard
├── SWARM_SYNC.md ← Agent handover log (Claude + Gemini)
├── SESSION_TRAIL.md ← Running session log
└── bin\ / cmdAlias\ ← Custom CLI tools on PATH
G:\JAX_PRIME\
├── 00_CORE\ ← K8s manifests, Vault
├── 01_ACTIVE\ ← Live scripts: swarm_dashboard_v9_secure.py, sovereign_forge.py, oracle_synthesis.py
├── SOVEREIGN_MESH\ ← n8n workflows, hydra_gateway, master_control, s22_edge_node
├── OPENCLAW_SWARM\ ← WSL golden images, spawn scripts
└── VAULT\ ← Knowledge vault (Hex vault intel docs)
| Service | Port | Location |
|---|---|---|
| Ollama | 11434 | Native Windows (GPU: RTX 5090) |
| LiteLLM gateway | 14000 | Docker (jax-prime-core) — port 4000 taken by Tailscale |
| n8n (jax-n8n) | 5680 | Docker — basic auth admin/sovereignty_2025_CHANGE_ME; owner: admin@jax.local/Sovereignty2025X; API key: see .env N8N_API_KEY (JWT, regen 2026-03-25) |
| n8n (strix_n8n) | 5678 | Standalone Docker — separate workflows at G:\JAX_PRIME\strix_compute\n8n_data |
| Open WebUI | 18080 | Docker — port 8080 taken by OpenClaw A2A broker |
| PostgreSQL | 5433 | Docker |
| Redis | 6379 | Docker |
| Nexus Dashboard | 5050 | Python (G:\JAX_PRIME\SOVEREIGN_MESH\master_control.py) |
| Pihole DNS | 53/80 | Docker |
Docker Compose: C:\Users\hjg42\SOVEREIGNTY\jax-prime-core\docker-compose.yml
cd C:/Users/hjg42/SOVEREIGNTY/jax-prime-core
docker compose up -dpython C:\Users\hjg42\JAX_SCRIPTS\JAX_REBOOT_WATCH.py # ADB connectivity monitor
python C:\Users\hjg42\JAX_SCRIPTS\JAX_NEURAL_CORE.py # Neural core agent
.\JAX_SCRIPTS\JAX_ARSENAL_LAUNCHER.ps1 # Launch Steam/Discord/NSight
.\JAX_SCRIPTS\JAX_DOCKER_WATCH.ps1 # Auto-deploy on Docker recovery- Note 10+ (SN: RF8MC0S7B2L) — ADB via USB, lockdown protocol in JAX_CLI\JAX_REBOOT_WATCH.py
- S22 (R5CTB2C858V) — Connected USB ADB, Ollama URL set to http://100.119.21.113:11434
- Firmware staging:
C:\Users\hjg42\ANDROID_TOOLS\(Frija/Odin/samloader) - FW binaries:
C:\Users\hjg42\FW_STAGING\
adb devices -l
adb -s R5CTB2C858V shell "..."| Node | Role | IP | OS |
|---|---|---|---|
| Strix Scar 18 (Crown) | Master compute | 100.119.21.113 | Windows 11 |
| m15r3 / Alienware (Vault) | Memory store | 100.68.114.40 | Windows (offline 19h) |
| ROG Ally (Scout) | Edge compute | 100.88.118.9 | Windows (relay sfo) — booted Windows not SteamOS |
| S22 (Watcher) | Mobile sensor | via hotspot | Android |
| jax-cloud-reflex | Cloud relay | 100.120.178.93 | OFFLINE |
- LAN: Strix=10.0.0.146, Xfinity=10.0.0.1, Android TV=10.0.0.114
- Hotspot: JAX-PRIME 5GHz, 4 clients — NAT persisted via
JAX-HotspotNAT-Persistscheduled task - Pihole DNS: Running in Docker, handles .jax.local resolution
- IronClaw:
.ironclaw/— connects tolocalhost:9999/v1(Hydra Gateway), 5 parallel jobs
- GWS (gws CLI): LIVE — Drive/Gmail/Sheets/Calendar via gcloud client 32555940559.
gws auth statusto verify. - GitHub CLI: Authenticated as HighTechHec
- Gemini CLI: hjg420@gmail.com — settings at
C:\Users\hjg42\.gemini\settings.json - Anthropic API: Set in Windows env +
.bashrc - Tailscale API key:
G:\JAX_PRIME\SOVEREIGN_MESH\.env
D:\GLOBAL_MEMORY_MANIFEST.md ← canonical mesh truth — READ THIS
D:\PROMETHEUS_PROTOCOL.md ← permanent cognitive anchor — READ BEFORE MAJOR TASKS
D:\EXOCORTEX_ENTERPRISE\ ← enterprise projects (Team_Kappa_YouTube, infra)
D:\UNIFIED_BRAIN_5.0\01_SOVEREIGN\logs\ ← sovereign brain logs
D:\SOVEREIGN_NEXUS\01_CODE\GLOBAL_CACHE\ ← pip-cache (233K files, exclude from index)
D:\SOVEREIGN_NEXUS\02_MEMORY\JaxPrime\ ← brain_vector_db, Chats, Google Keep exports
D:\Technoticia_Nexus\ ← PARA knowledge (Inbox/Projects/Areas/Resources)
- Location:
C:\Users\hjg42\JAX_BRAIN_INDEX.db— 1.1GB FTS5 SQLite - Indexer:
C:\Users\hjg42\JAX_SCRIPTS\JAX_NEXUS_INDEXER.py— 278,580 files indexed - MESH_MAP:
C:\Users\hjg42\MESH_MAP.md+G:\JAX_PRIME\MESH_MAP.md - Query:
SELECT path,preview FROM files WHERE tags LIKE '%todo%'or use FTS:SELECT * FROM files_fts WHERE files_fts MATCH 'ollama docker'
- On session start: Read
SWARM_SYNC.md+D:\PROMETHEUS_PROTOCOL.mdfor current mesh state - During session: Append actions to
SESSION_TRAIL.md - On handoff: Update HANDOVER LOG in
SWARM_SYNC.md
- Platform: Windows 11, shell = bash (Git Bash). Use Unix paths unless calling
.ps1directly. - Python: prefer
pip install --user - msr.exe / nin.exe — custom search binaries in home root
- G: drive is the active backbone (F: is dead — never reference F:\JAX_PRIME or F:\MYCELIUM_HIVE)