- Voice commands (#3): say "restart the homebridge agent" or "restart mediastack agent" → queues restart_service to that agent; lists options if no hostname matched; 6 KB intents added
- Persistent context (#4): chat.php now loads last 6 turns from most recent prior session before current session history, giving JARVIS memory across page reloads
- Proactive reminders (#5): 3s after login, auto-announces overdue tasks / tasks due today / upcoming appointments; 5-min interval checks for appointments starting within 15 min and speaks alert once
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Two issues:
1. NM_OPEN_RE regex was too narrow — phrases like "show me the network",
"open the network", "show network status" did not match, so they fell
through to the API which returned text but never opened the map.
Broadened regex to catch natural network-related phrases.
2. When the network_scan intent IS triggered via API, the map never opened
because the API response handler only processed reply/arc_job.
chat.php now returns open_network_map:true for network_scan intent,
and the client calls openNetMap() when that flag is present.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
After writing the updated script, _stop_event.set() caused a clean exit
(code 0). SCM failure recovery only fires on non-zero exit, so the service
stayed down permanently after every auto-update.
Fix: set _update_restart=True before signalling stop; SvcDoRun() checks
the flag after main() returns and calls sys.exit(1), which triggers the
sc.exe failure recovery chain (restart/5s/10s/30s configured at install).
The LocalSystem service account cannot access per-user Python installs
(AppData\Local\Programs\Python\...). When a per-user install is detected,
automatically install Python system-wide before proceeding.
- Detect per-user Python (AppData in path) and trigger system-wide install
- Extract system Python install logic into Install-PythonSystemWide function
- Check winget exit code before marking install successful
- Split pip install and postinstall into separate steps; pip failure is fatal,
postinstall failure is a warning (service DLLs may already be registered)
- Use $LASTEXITCODE check on pip rather than try/catch (external process)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace the scheduled-task approach (required user to stay logged in) with a
proper Windows Service using pywin32. The service runs as LocalSystem, starts
at boot, and auto-restarts on failure — no PowerShell window needed.
Agent changes (jarvis-agent-windows.py):
- Add Windows Service class via pywin32 (JarvisAgentService)
- Cleanly handles SvcStop by setting a threading.Event
- main() loop uses _stop_event.wait() instead of time.sleep() so stop is immediate
- self_update() signals the stop event when running as a service (SCM restarts it)
- __main__ block dispatches to SCM entry point or HandleCommandLine (install/stop/remove)
- Falls back to direct run if pywin32 not installed (for debugging)
Installer changes (install-windows.ps1):
- pip install pywin32 + postinstall (registers service runner DLLs)
- Python search prefers system-wide install (accessible by LocalSystem)
- Downloads Python 3.11 directly from python.org for Win 8.1 machines without winget
- Removes legacy JARVIS-Agent scheduled task if present
- Registers JARVISAgent service with --startup auto
- Configures sc.exe failure recovery (restart at 5s/10s/30s)
- Updated management commands in summary (Start-Service, Stop-Service, etc.)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add version column to registered_agents table
- Agents send version on registration (Linux 3.1, Windows 3.0, macOS 3.0)
- workers_list API returns latest_versions per platform
- Workers tab: VERSION column with green check (up-to-date) or red (outdated)
- Outdated agents highlight row and show blue UPDATE button
- Up-to-date agents show dimmed UPDATE button
- Update button dispatches update command immediately
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Old installer wrote to /opt/jarvis-agent/config.json with server_url/api_key/
heartbeat_interval keys and pre-registered with JARVIS. Agent v3.0 expects:
- Config at /etc/jarvis-agent/config.json with jarvis_url/registration_key/
hostname/poll_interval/heartbeat_every keys
- api_key stored by agent itself in /var/lib/jarvis-agent/state.json
- Agent self-registers at startup using registration_key
Also adds: imagemagick install (headless screenshot support), apk support
for Alpine/WireGuard, copies to /usr/local/bin/jarvis-agent.py.
Previously agents only registered when api_key was missing (first run).
After updating to v3.0 with screenshot capability, restarted agents never
refreshed their capabilities in the DB. Now register() is called every
startup — server does UPDATE on existing agent_id so api_key is preserved.
self_update(cfg) and shell allow_shell_commands check both referenced cfg
from run() scope, but execute_command() is a standalone function. Fixed by
calling load_config() locally in each branch that needs it.
A ParseError or fatal in any endpoint file now returns JSON 500 for that
endpoint only. switch replaced with data-driven map. All other endpoints
continue working normally when one is broken.
Arc Reactor was running from /opt/jarvis-arc with no version tracking.
Added to deploy/ so all fixes (metrics_json→metric_data, flat JSON parsing,
disk mount key fix) are captured. WG configs are runtime-generated secrets
and must not be committed.
Swap: replace grid-column reassignment with named grid-template-areas
(grid-column approach caused leftPanel to disappear and mis-position).
"left center right" <-> "right center left" cleanly swaps both columns.
Collapsible: clicking any panel-title collapses/expands that panel.
Chevron rotates -90deg when collapsed. State persists in localStorage.
All interactive elements in title bars (+ buttons etc) still work.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Broke entire JS block — side panels, data loading, everything.
synthesis.length>1500 ternary had bare newlines inside single quotes.
Replaced with \n escape sequences.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
const ts = ts() in loadVision caused TDZ ReferenceError crashing gallery.
visionRunScreenshot now fetches online agents from agents_list API when
no screenshots exist yet (previously showed No agents online falsely).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
ARC REACTOR tab had HTML and PHP API handlers but no JS load function,
causing ReferenceError on every tab click. Adds all three missing functions.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Rename memory section CAT_COLORS to MEM_CAT_COLORS to avoid SyntaxError
that prevented doLogin from being defined, locking users out of admin.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- DB: missions, mission_steps, mission_runs tables
- reactor.py v7.0.0: handle_run_mission, _execute_mission, mission_trigger_loop (schedule/guardian_event/email_keyword triggers), {{template}} substitution across steps, full CRUD REST endpoints
- arc.php: missions/mission_get/mission_runs/mission_create/mission_update/mission_delete/mission_run/mission_toggle actions
- admin/index.php: Mission Ops tab with visual workflow builder (trigger config, step cards with ↑↓, JSON payload editor, continue-on-failure flag), run history with step-level detail, enable/disable toggle
- index.html: MISSIONS tab with collapsible mission cards, RUN NOW button per mission, live run result feedback
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- chat.php: Add Tier 0.9a (gmail_triage), Tier 0.9b (remote_exec) detection;
refactor arc submit into arcSubmitJob() helper; natural-language triggers for
email triage (check my email, triage inbox) and remote exec (restart X on Y,
run X on Y, get logs from X on Y)
- arc.php: Add triage and triage_action endpoints (read/update email_triage table)
- index.html: Add COMMS tab with triage card UI (filter bar, category badges,
draft reply viewer, copy/dismiss actions); loadComms() with 8s polling;
onArcJobStarted() routes gmail_triage jobs to COMMS tab
- admin/index.php: Add GMAIL TRIAGE section under COMMUNICATIONS nav; triage_list/
triage_action/triage_run PHP actions; loadTriage() JS with full table + draft
modal; triageRunNow() submits gmail_triage job to Arc Reactor
Arc Reactor v2.0:
- research handler: DDG search → async page fetch → trafilatura extraction → Claude synthesis
- tool_loop handler: multi-step agent loop (up to 200 iter) with web_search, fetch_url, jarvis_agents, jarvis_alerts, current_time tools
- llm handler: multi-provider router (Claude/Groq/Ollama)
- /jobs/recent endpoint for HUD polling
- Phase 1 handlers preserved (ping/echo/shell)
chat.php — Tier 0.9 Intel Protocol (before KB intent engine):
- Detects: research/investigate/deep-dive/look up/find out about → research job
- Detects: step-by-step/figure out/analyze and report → tool_loop job
- Returns arc_job ID in response for UI polling
- Depth modifiers: quick/standard/deep
index.html:
- INTEL tab in right panel tab bar
- Research result cards with expand/collapse, synthesis, sources, status
- Live polling (4s) when INTEL tab is active + active jobs present
- Auto-switches to INTEL tab when research is triggered from chat
- intelPrompt() pre-fills chat input for new research
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Python asyncio daemon (/opt/jarvis-arc/reactor.py) running on 127.0.0.1:7474
- systemd service (jarvis-arc) auto-starts with MySQL dependency
- arc_jobs + arc_status MySQL tables for async job queue
- api/endpoints/arc.php: PHP bridge to daemon (status, job_create, job_get, jobs, purge)
- api.php: added arc route
- index.html: ARC REACTOR status indicator in bottom bar with live polling
- admin/index.php: ARC REACTOR nav section + full job management panel
- Built-in job handlers: ping, echo, shell (whitelist-gated)
- Foundation for Phase 2 (Intel Protocol) and beyond
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Replace tiny dot nodes with frosted-glass bubbles with ambient glow and float animation
- Add 6th ring for netscan-discovered network devices (cap 28)
- Split named/DB devices and discovered devices into separate rings
- Push rFrac to 0.82 to fill the overlay window
- Increase all ring caps and node radii
- Add FortiGate NAT IP to providers ACL
- Fix TCP SIP drop issue via transport=udp
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Root cause 1: Notification.requestPermission() on startup (3s delay) opened
a browser permission dialog while JARVIS was still speaking the welcome greeting.
This aborted the SpeechRecognition session. Because isSpeaking=true at that
moment, onend did not reschedule a restart — mic went permanently silent.
Fix: removed the startup requestPermission() call entirely.
Root cause 2: Same requestPermission() inside _focusWindow() called on every
enterVoiceMode() — could abort the recognition session on each wake.
Fix: only create notification when permission already granted, never request.
Root cause 3: SLEEP_CMDS matched bare words like offline and sleep that appear
in normal commands (check if server is offline, put device to sleep, etc.)
Fix: tightened to require explicit phrasing — go offline, sleep mode,
shut down jarvis, good night jarvis, etc.
Sleep mode:
- Commands: good night / go to sleep / shut down / standby / offline / signing off
- Works via voice or text
- UI dims to 8% brightness, slow-spinning standby reactor overlay appears
- Refresh loop pauses (light 2min heartbeat keeps session alive)
- Mic stays fully active — only responds to master wake phrases
- Idle-reload disabled while sleeping (prevents unwanted reloads overnight)
Wake from sleep (master wake phrase):
- Detects wake phrase while isAsleep=true, routes to wakeFromSleep()
- Full HUD boot sequence animation (panels slide in)
- refreshAll() fires immediately to reload all data
- JARVIS greets: All systems back online
Window focus on any wake:
- window.focus() called on every enterVoiceMode
- document.title flashes 8x between JARVIS ONLINE and default
- Web Notifications API: system popup fires when window is minimized/backgrounded
- Notification permission requested 3s after login
- Works regardless of sleep/voice mode
Reverted to c8e0020 (all 10 effects working) then added net map cleanly:
- Used var/for instead of const/let/arrow-functions to avoid any closure/scope issues
- Orbital ring layout: JARVIS hub center, 4 concentric rings (proxmox/services/agents/devices)
- Rings rotate at different speeds/directions independently
- Spoke lines hub-to-each-node with cyan inbound and orange outbound particles
- Node labels point outward from center, never overlap
- Tiny green/red status dot on every non-hub node
- Hover shows node info card (name/IP/status/ring)
- Open: say/type show network map / network topology / show connections
- Close: say/type close map / close network / dismiss map
- All other features (mic, voice, text, panels) unaffected
- Overlay shrunk to min(860px, 86vw) x min(570px, 76vh) — fits any screen
- Solar system layout: JARVIS hub at center, 4 concentric orbital rings
- Ring 1 (green, innermost): Proxmox nodes, rotates clockwise
- Ring 2 (gold): Services (HA, AI, PBX, Homebridge), counter-clockwise
- Ring 3 (cyan): Linux/Windows agents, slow clockwise
- Ring 4 (dim blue, outermost): Network scan devices, slow counter-clockwise
- Rings rotate at different speeds/directions (persistent offsets across frames)
- Each ring draws dashed track + tick marks every 30 deg + label + online/total count
- Nodes evenly spaced on their ring, rotate with it
- Spoke lines from each node to hub (straight, low opacity)
- Particle flow on spokes: cyan in (data), orange out (commands)
- Node positions computed per-frame from angle+ring radius+rotation
- Overlay canvas sized exactly to container minus header+legend height
- Quadrant layout: Proxmox (top-left), Services (top-right), Agents (bottom-left), Devices (bottom-right)
- Dashed divider lines + subtle per-zone color gradient fill separates sections visually
- Zone watermark labels (PROXMOX CLUSTER, SERVICES, AGENTS, NETWORK DEVICES) with online/total count
- Nodes arranged in tidy grid within each zone — no more single crowded ring
- Labels positioned OUTWARD from hub center (atan2 to calculate angle) so they never overlap nodes
- Bezier lines bow outward away from hub center (control point pushed along hub→midpoint vector)
so lines spread out and each is individually traceable
- IP shown only on hub and hovered nodes — reduces label clutter
- Overflow indicator: shows +N MORE when zone has too many nodes (max per zone: 6/6/8/10)
- Intra-zone cross-links for Proxmox cluster (green) and Services cluster (gold)
- RGB color system replaces old r/g/b object — cleaner rgba() template strings
- Voice: say show network map / network topology / show connections to open
- Voice: say close map / dismiss / close network to close
- Same commands work in chat text input
- Explode animation: overlay expands from top-left reactor position with clip-path wipe
- Collapse animation: folds back to reactor on close
- Visualization: live node graph with bezier curved edges, hub (JARVIS) at center
- Inner ring: all registered agents (agents color-coded by type: proxmox=green, HA=gold, etc)
- Outer ring: netscan-discovered devices
- Rotating orbit rings on hub and agent nodes
- Pulsing radial glow per node keyed to online status
- Hub cross-hair targeting lines
- Directional particle flow:
- CYAN particles: data/heartbeats flowing FROM agents TO JARVIS hub
- ORANGE particles: commands flowing FROM JARVIS hub TO agents
- All particles travel curved bezier paths, fade at endpoints, glow with shadows
- Mouse hover: node info card shows name/IP/status/type
- Stats bar: total nodes, online count, agent count
- Background: faint hex grid overlay for sci-fi depth
myron