18 KiB

Raw Blame History

TOOLS.md - Local Notes

Skills define how tools work. This file is for your specifics — the stuff that's unique to your setup.

James Server Hardware (forge — 192.168.1.16)

CPU: Intel i7-6700K @ 4.0GHz (4c/8t)
RAM: 64GB DDR4
GPU: NVIDIA GTX 970 4GB
Storage: 477GB NVMe (Samsung 950 PRO 512GB)
OS: Ubuntu 24.04.1 LTS headless + minimal GUI for headed Chrome
Hostname: forge

See memory/infrastructure.md for full infrastructure map.

What Goes Here

Things like:

Camera names and locations
SSH hosts and aliases
Preferred voices for TTS
Speaker/room names
Device nicknames
Anything environment-specific

Examples

### Cameras
- living-room → Main area, 180° wide angle
- front-door → Entrance, motion-triggered

### SSH
- home-server → 192.168.1.100, user: admin

### TTS
- Preferred voice: "Nova" (warm, slightly British)
- Default speaker: Kitchen HomePod

James Dashboard

URL: http://100.123.216.65:9200 (Tailscale) or http://localhost:9200
Purpose: Visual status board for tasks, briefings, history

Tasks API:

GET /api/tasks - list all tasks
POST /api/tasks - add task
PATCH /api/tasks/:id - update task
DELETE /api/tasks/:id - remove task
Fields: title, text, priority, status, owner, domain, notes
Priority: high, medium, low
Status: pending, in-progress, done
Owner: "johan" or "james" (who owns the task)
Domain: Kaseya, inou, Sophia, ClawdNode, Infrastructure, Personal, etc.

News API:

GET /api/news - list news (newest first, max 20)
POST /api/news - add news item
DELETE /api/news - clear all news
DELETE /api/news/:id - remove specific item
Fields: title, body, type, source (optional), url (optional)
Type: info, success, warning, error

Briefings API:

GET /api/briefings - list briefings (newest first, last 30)
POST /api/briefings - add briefing
GET /api/briefings/:id - get single briefing
DELETE /api/briefings/:id - remove briefing
Fields: title, date, weather, markets, news, tasks, summary

Deliveries API:

GET /api/deliveries - list active deliveries (excludes delivered)
GET /api/deliveries?all=true - list all including delivered
POST /api/deliveries - add delivery (prefer upsert instead)
PUT /api/deliveries/upsert - smart upsert: matches by tracking_number or description+retailer, updates existing or creates new. Always use this for shipping email triage.
GET /api/deliveries/:id - get single delivery
PATCH /api/deliveries/:id - update delivery
DELETE /api/deliveries/:id - remove delivery
Fields: description, carrier, retailer, tracking_number, tracking_url, expected_date, status, notes
Status values: shipped, in_transit, out_for_delivery, delayed, delivered

Status API:

GET /api/status - list all status items
POST /api/status - set status {"key": "...", "value": "...", "type": "info|warning|error"}
GET /api/status/:key - get single status
DELETE /api/status/:key - remove status
Use for: Claude usage, system health, key metrics displayed at top of dashboard

Workflow:

Morning brief → POST to /api/briefings, send Signal link to dashboard
Track tasks with owner field (mine vs Johan's)
Keep briefing history for reference
Update Claude usage status: scripts/claude-usage-check.sh (auto-updates dashboard)

Forge = James' Home (192.168.1.16)

forge IS the James server now. See hardware section above.
GPU: GTX 970 4GB (Driver 580.126.09, CUDA 13.0)
Ollama: Installed (0.15.4)
Python: /home/johan/ocr-env/ (venv with PyTorch 2.2 + CUDA 11.8)
SSH: Key auth only (password auth disabled)
Firewall: UFW active, SSH + LAN allowed
Owner: James ⚡ (full autonomy)

OCR Service (GLM-OCR):

URL: http://192.168.3.138:8090
Service: systemctl --user status ocr-service (on forge)
Source: /home/johan/ocr-service/server.py
Model: /home/johan/models/glm-ocr (zai-org/GLM-OCR, 2.47 GB)
VRAM usage: 2.2 GB idle (model loaded), peaks ~2.8 GB during inference
Performance: ~2s small images, ~25s full-page documents (auto-resized to 1280px max)
Endpoints:
- GET /health — status + GPU memory
- POST /ocr — single image OCR (multipart: file + prompt + max_tokens)
- POST /ocr/batch — multi-image OCR
Auto-resize: Images capped at 1280px longest edge (prevents OOM on GTX 970)
Usage from james: curl -X POST http://192.168.3.138:8090/ocr -F "file=@image.png"
Patched env: transformers 5.0.1.dev0 (from git) + monkey-patch for PyTorch 2.2 compat

Home Network

Public IP: 47.197.93.62 (not static, but rarely changes)
Location: St. Petersburg, Florida
Caddy (reverse proxy): 192.168.0.2 / Tailscale: 100.84.42.55 (caddy)
- SSH: tailscale ssh root@caddy or ssh root@caddy (key installed)
- Config: /etc/caddy/Caddyfile

James Server (Hetzner)

LAN IP: 192.168.1.16
Gateway UI: http://192.168.1.16:18789/
Agents:
- Main (James): http://192.168.1.16:18789/
- Mail Agent: http://192.168.1.16:18789/agents/mailagent

Uptime Kuma

URL: http://zurich.inou.com:3001
User: james
Password: WW8ipJfY27ELf7nnouaKLCL6

Openprovider (Domain Registrar)

URL: https://cp.openprovider.eu
User: johan.jongsma@iasobackup.com
Password: !!Helder06

Test Devices

ThinkPhone 1 (Motorola/Lenovo) — Johan's Android test device

Git Server (Zurich)

Host: zurich.inou.com
User: git
URL format: git@zurich.inou.com:<repo>.git
Repos: azure-backup, clawdnode-android, inou-mobile, mail-agent
Auth: SSH keys (claude@macbook, james@server, johan@ubuntu2404 authorized)
Create new repo: ssh git@zurich.inou.com "git init --bare <name>.git"

myCigna (Health Insurance Portal)

URL: https://my.cigna.com
Username: tjjongsma
Password: QFL&ARHXGW4R
2FA: Email to tj@jongsma.me (I can grab the code from MC) or SMS to ***-2475
Account holder: Tatyana
Covered: Tatyana, Johan, Michael, Sophia
Access method: Real Chrome on Xvfb:99, CDP on port 9224 (headless Playwright gets WAF-blocked)

SSH Hosts

hostkey50304 → 82.22.36.202 / zurich.inou.com (CH/Switzerland VPS)
- Location: Zürich, likely Equinix ZH (Josefstrasse 225)
- Upstream: Cogent Communications
- Specs: 4 vCore, 6GB RAM, 120GB SSD
- OS: Fresh install (2025-01-27)
- User: root
- Purpose: Security infrastructure (geographic diversity for monitoring, SOC2 compliance). NOT for hosting inou.com.
Home Assistant → 192.168.1.252
- User: root
- Port: 22
- ⚠️ STRICT RULES:
  - NO changes without Johan's explicit permission
  - NEVER change lights during night hours
  - NEVER play audio on speakers/tablets during night hours
  - Night = Sophia's sleep/care time — disruptions are dangerous

Browser Profiles

Profile	Type	Port	Use Case
chrome	Relay (your actual Chrome)	18792	X.com, paranoid sites, full auth
fast	Headless Chromium	9223	General automation, tolerant sites
clawd	Headless (managed)	18800	Default, no auth

Chrome Relay (profile="chrome") — Best for authenticated sites

Attaches to your actual Chrome browser via extension. No session copying, no detection issues.

Setup (one-time):

Extension installed at ~/.clawdbot/browser/chrome-extension
Load in Chrome: chrome://extensions → Developer mode → Load unpacked
Pin the extension to toolbar

Usage:

Navigate to site in Chrome, make sure you're logged in
Click the Clawdbot extension icon (badge shows ON)
I use browser(profile="chrome") to control that tab

When to use Chrome Relay:

X.com (very aggressive bot detection)
Sites requiring 2FA/login you've already completed
Anything that blocks headless browsers

Headless Chromium (profile="fast") — For general automation

Runs headless with synced cookies from Chrome. Some paranoid sites (X.com) detect and block it.

Setup script: ~/clawd/scripts/browser-setup.sh

scripts/browser-setup.sh start-login    # Open Chrome, login to sites
scripts/browser-setup.sh sync           # Close Chrome first! Then sync
scripts/browser-setup.sh start-headless # Start headless on port 9223
scripts/browser-setup.sh status         # Check what's running
scripts/browser-setup.sh stop           # Stop all

⚠️ Important: Close Chrome before running sync — copying a live profile invalidates sessions!

Usage: browser(action="...", profile="fast")

Browsing Tips

Use web_fetch for simple page reads (faster, cheaper than full browser)
Use browser when you need JavaScript rendering, interactions, or auth
For large pages, use maxChars on snapshots to avoid context bloat
Keep targetId from snapshot responses for stable tab references

bird (X/Twitter CLI)

Wrapper: ~/clawd/scripts/bird (sets auth tokens automatically)
Config: ~/.config/bird/config.json5 (tokens stored but not read properly - use wrapper)
Usage: ~/clawd/scripts/bird read <tweet-id> or bird read <url>
Commands: bird search, bird home, bird mentions, bird news, bird user-tweets @handle
For X.com access — use bird instead of browser (faster, no bot detection issues)
Auth: Using @johanjongsma account

James Email Identity

Address: james@jongsma.me (also james@inou.com)
Password: JamesCoS2026!
IMAP: mail.jongsma.me:993 (TLS)
SMTP: mail.jongsma.me:587 (STARTTLS)
Stalwart admin: admin / JamesAdmin2026x (http://127.0.0.1:8880 on Zurich)
Use for: Outbound email as James, direct comms with Tanya/family

Proton Mail Bridge (Headless)

Service: systemctl --user status protonmail-bridge
Account: tj@jongsma.me (Tanya & Johan Jongsma)
IMAP: 127.0.0.1:1143 (STARTTLS)
SMTP: 127.0.0.1:1025 (STARTTLS)
Bridge Password: BlcMCKtNDfqv0cq1LmGR9g
Keychain: pass (no gnome-keyring needed)
Config: ~/.config/protonmail/bridge-v3/prefs.json → {"preferred_keychain": "pass"}

Message Bridge (WhatsApp backend) - whatsmeow

Service: systemctl --user status message-bridge
Port: 8030
Source: /home/johan/dev/message-bridge/
Data: ~/.message-bridge/
Linked Number: +1 727 225 2475 (Johan's ThinkPhone)
Note: MC wraps this — use MC API for WhatsApp, not this directly

Direct API (for debugging):

GET /status - connection status
GET /messages - list messages
GET /qr?format=png - QR code for linking (if disconnected)
POST /send - send message

Message Center (MC) - Unified Messaging API

Service: systemctl --user status mail-bridge
Port: 8025
Source: /home/johan/dev/mail-bridge/ (branch: mc-unified)
Data: ~/.message-center/ (cursors.json, orchestration.db)
Health: curl http://localhost:8025/health

Connectors:

tj_jongsma_me — IMAP (tj@jongsma.me via Proton Bridge)
johan_jongsma_me — IMAP (johan@jongsma.me via Proton Bridge)
whatsapp — HTTP wrapper for message-bridge on port 8030

Unified API:

GET /messages/new — unseen messages from all sources
GET /messages?since=24h — replay window
GET /messages/{id} — single message
POST /messages/ack — advance consumer cursor

Actions:

POST /messages/{id}/archive — mark seen/archive
POST /messages/{id}/delete — delete message
POST /messages/{id}/reply — send reply {"body": "..."}
POST /messages/{id}/to-docs — forward attachments to ~/documents/inbox/

Orchestration DB: ~/.message-center/orchestration.db

Tracks: first_seen, last_action, acked_by per message
Actions persist across restarts

Webhook: POSTs {"event": "new"} to http://localhost:18789/hooks/messages

Telegram Bot (@jamesjongsma_bot)

Bot: @jamesjongsma_bot → t.me/jamesjongsma_bot
Bot ID: 8510971070
Token: 8510971070:AAFFgv_UO_9L0Ulp2DRKHD-IWKkrarJNTIc
Owner account: @johanjongsma (Telegram ID: 8454563068)
Channel config: channels.telegram in openclaw.json, dmPolicy: open, allowFrom: ["*"] (required — open policy won't start without explicit allowFrom)
Status: Live as of 2026-02-18
Note: Created via BotFather from Johan's @johanjongsma account

Commands

/screenshot → Pull latest screenshot from Mac desktop (uses screenshot skill)

Uptime Kuma (Zurich)

URL: https://kuma.inou.com (DNS → zurich 82.22.36.202, Caddy reverse proxy → localhost:3001)
User: james / JamesKuma2026!

Env vars for kuma.py:

UPTIME_KUMA_URL=http://localhost:13001  # (via SSH tunnel: ssh -L 13001:127.0.0.1:3001 root@zurich.inou.com)
UPTIME_KUMA_USERNAME=james
UPTIME_KUMA_PASSWORD=JamesKuma2026!

Python venv: /home/johan/clawd/skills/uptime-kuma/.venv/
Notification channels:
- [1] Clawdbot Signal Alert (webhook → OC) — for MC alerts
- [2] Johan Direct (ntfy) — for OC/network alerts
Monitors:
- [1] inou.com HTTP
- [2] inou.com API
- [3] Forge — OpenClaw (push, token: 1ae2c3adb6)
- [4] Forge — Message Center (push, token: 2fe6898f36)

ntfy (Zurich — self-hosted)

URL: https://ntfy.inou.com (Caddy → localhost:2586)
User: james / JamesNtfy2026!
API Token: tk_ggphzgdis49ddsvu51qam6bgzlyxn
Alert topic: forge-alerts (anonymous read allowed for iOS app)
Johan subscribes to: https://ntfy.inou.com/forge-alerts (in ntfy iOS app)

ntfy (Push Notifications)

Server: https://ntfy.inou.com (self-hosted on Zurich)
Auth: Authorization: Bearer tk_k120jegay3lugeqbr9fmpuxdqmzx5 (james admin account)
Markdown: Supported — use -H "Markdown: yes" header
Headers: Title, Priority (1-5), Tags (comma-separated emoji shortcodes)

Topics:

forge-alerts — OC/infrastructure alerts (when Signal/OC is down). Anonymous read.
inou-alerts — inou health platform alerts. Use instead of Signal for inou-related notifications. Anonymous read.

Use inou-alerts for: inou app errors, health data anomalies, inou server issues Use forge-alerts for: OC down, MC down, forge/network issues

curl -s https://ntfy.inou.com/inou-alerts \
  -H "Authorization: Bearer tk_k120jegay3lugeqbr9fmpuxdqmzx5" \
  -H "Title: inou alert" \
  -H "Tags: warning" \
  -H "Markdown: yes" \
  -d "Your message"

Johan subscribed on: iPhone 17 (migrating as of Feb 19 2026), was Android

Health Push Script

Script: /home/johan/scripts/health-push.sh
Cron: * * * * * (every minute)
Logic: Checks MC health + OC health locally, pushes to Kuma only if healthy
Alert routing:
- MC down → James (OC webhook) — James can investigate
- OC down → Johan direct (ntfy) — James IS the thing that's down
- Home network down → Johan direct (ntfy) — everything at home is down

OpenProvider API (Domain Registrar)

API: https://api.openprovider.eu/v1beta/
Creds: /home/johan/.config/openprovider.env — OP_USERNAME, OP_PASSWORD
Auth: POST /auth/login → bearer token
NS change: PUT /domains/{id} with {"name_servers":[...]}
List domains: GET /domains
I have full access — don't ask Johan to change NS manually!

Why Separate?

Skills are shared. Your setup is yours. Keeping them apart means you can update skills without losing your notes, and share skills without leaking your infrastructure.

Add whatever helps you do your job. This is your cheat sheet.

Govee H5122 Buttons

Button 1: event.gv5122775b_button_1 (MAC: D2:2D:83:86:77:5B)
- Automation: automation.govee_button_mbed_pendants_toggle → toggles switch.mbed_pendants
Button 2: event.gv51222839_button_1 (MAC: D2:2D:80:C6:28:39) — Office
- Automation: automation.govee_button_2_suction_machine → toggles suction machine
Pairing: Hold 5 sec (LED flashes), press once to broadcast, add from HA UI quickly
Note: BLE proxy must be enabled on nearby Athom sensor (Office1 b372f4 has it now)

Office Tablet (office1.tbl)

Media player: media_player.lenovo_tab_m8_7
TTS notify: notify.lenovo_tab_m8_text_to_speech_7
Overlay notify: notify.lenovo_tab_m8_overlay_message_7
Screen: light.office_tbl_screen
Fully Kiosk media_player: media_player.office_tbl
Use for: James voice output testing, announcements

Spacebot / Andrew (192.168.1.17 — parallel test alongside OpenClaw)

Container: spacebot on 192.168.1.17, port 19898, restart=unless-stopped
Web UI: http://192.168.1.17:19898
Config: /home/johan/spacebot-config.toml (mounted to /data/config.toml)
Telegram bot: @Andrew_Jongsma_bot (Johan named it "Andrew")
Telegram token: 8737175907:AAEcT0Pjqne46rZhRtZsR8hdRgoKicR5TCo
LLM: Claude Sonnet 4.6 via Anthropic OAuth — switched 2026-02-24
Auth: OAuth stored in /data/anthropic_oauth.json (Johan's claude.ai account)
Routing: channel/branch/worker = anthropic/claude-sonnet-4-6, cortex/compactor = anthropic/claude-haiku-3-5
Fireworks key: fw_RVcDe4c6mN4utKLsgA7hTm (kept as fallback, not currently used)
Known issue: channel model calls reply() and stops — worker dispatch announced but never executed. Simple Q&A works; multi-step agentic tasks silently fail. Revisiting 2026-03-03.
Memory ingest: /home/johan/spacebot-ingest/ → mounted at /data/agents/main/workspace/ingest
Binding: chat_id 8454563068 (Johan's Telegram) → agent main
Note: bird CLI = steipete's xurl skill — already using it, no Chrome Extension relay needed

Fireworks AI

API Key: fw_RVcDe4c6mN4utKLsgA7hTm ⚠️ (fw_TGADpSki7zak4K9JxPzbXU is INVALID — expired)
Base URL: https://api.fireworks.ai/inference/v1
Privacy: Guaranteed zero retention — safe for private/medical data
Key models:
- accounts/fireworks/models/minimax-m2p5 — MiniMax M2.5 (230B MoE, fast, coding+agentic, 200K ctx)
- accounts/fireworks/models/kimi-k2p5 — Kimi K2.5 (multimodal, vision+text)
- accounts/fireworks/models/llama-v3p1-70b-instruct — Llama 3.1 70B
OpenClaw sessions_spawn: model="fireworks/accounts/fireworks/models/minimax-m2p5"

18 KiB Raw Blame History