πŸ€– EOS Agent Command Center

How your AI agents work, remember, and collaborate β€” a guide for humans

⚑ The 30-Second Version

You have three AI agents running 24/7 on Mac Minis. They can read files, send messages, check email, browse the web, and work on projects. But they have no permanent memory unless they write things down.

Key Insight: Think of each agent like an employee who gets amnesia every few hours but has an excellent filing system. The quality of their work depends entirely on how well they maintain their files.

Your Three Agents

πŸ’Ό
Valentina
Executive Assistant & Operations Lead

Email triage, scheduling, Zoho CRM, QuickBooks, V-PEG pricing engine, customer communications, team coordination

Machine
Mac Mini M4 Pro 64GB
Model
Claude Opus 4
Tailscale
100.87.49.18
Channel
iMessage
πŸ”§
Alex
Technical & Engineering

V-PEG drawings, technical integrations, PEG system analysis, engineering documentation, code development

Machine
Mac Mini M4 Pro 24GB
Model
Claude Opus 4
Tailscale
100.81.25.123
Channel
iMessage
🎨
Sage
Marketing & Social Media

Content creation, social media, competitive analysis, website content, brand strategy, SEO

Machine
Mac Mini (TBD)
Model
Claude Opus 4
Tailscale
100.99.253.24
Channel
iMessage

πŸ—οΈ System Architecture

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ YOUR PHONE β”‚ β”‚ (iMessage / Signal) β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚ β”‚ β”‚ β”Œβ”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β” β”‚ VALENTINA β”‚ β”‚ ALEX β”‚ β”‚ SAGE β”‚ β”‚ Mac Mini │◄───►│ Mac Mini │◄───►│ Mac Mini β”‚ β”‚ (Ops/EA) β”‚ SSH β”‚ (Tech) β”‚ SSH β”‚ (Mktg) β”‚ β””β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”˜ β”‚ β”‚ β”‚ β”Œβ”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β” β”‚ MEMORY.md β”‚ β”‚ MEMORY.md β”‚ β”‚ MEMORY.md β”‚ β”‚ memory/*.md β”‚ β”‚ memory/*.md β”‚ β”‚ memory/*.md β”‚ β”‚ AGENTS.md β”‚ β”‚ AGENTS.md β”‚ β”‚ AGENTS.md β”‚ β”‚ SOUL.md β”‚ β”‚ SOUL.md β”‚ β”‚ SOUL.md β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚ β”Œβ”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ SHARED SERVICES β”‚ β”‚ Zoho CRM β”‚ Gmail β”‚ QuickBooks β”‚ β”‚ Google Drive β”‚ Calendar β”‚ Strava β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

🧠 How Agent Memory Actually Works

This is the most important thing to understand. Agents don't have persistent memory. Every new conversation session starts completely blank.

What Happens When a Session Starts

Session Start β”‚ β–Ό β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ Load workspace files: β”‚ β”‚ β€’ MEMORY.md β”‚ ← Main "brain" file (always loaded) β”‚ β€’ AGENTS.md β”‚ ← Behavior rules β”‚ β€’ SOUL.md β”‚ ← Personality β”‚ β€’ IDENTITY.md β”‚ ← Who am I? β”‚ β€’ USER.md β”‚ ← Who are you? β”‚ β€’ HEARTBEAT.md β”‚ ← Proactive checklist β”‚ β€’ TOOLS.md β”‚ ← Environment notes β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚ β–Ό β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ That's it. That's all β”‚ β”‚ the agent "remembers" β”‚ β”‚ β”‚ β”‚ Everything else needs β”‚ β”‚ to be READ from files β”‚ β”‚ in the memory/ folder β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
The Repeat Work Problem: If an agent solves a hard problem but doesn't write the solution to a file, the next session will solve it again from scratch. This is why Alex kept redoing the logo fix β€” the solution wasn't saved in a status file the first time.

The File Hierarchy

FilePurposeWhen Updated
MEMORY.mdHigh-level facts, key decisions, system status. Always loaded on startup.When major things change
memory/YYYY-MM-DD.mdDaily work logs. Raw notes of what happened.During/after each session
memory/project/STATUS.mdProject checkpoints. What's done, what's next, solved problems.End of every work session
AGENTS.mdBehavior rules and standard procedures.When you want to change how the agent works
SOUL.mdPersonality and communication style.Rarely β€” it's the agent's core identity
TOOLS.mdEnvironment-specific notes (passwords, IPs, folder paths).When setup changes

The Status File Pattern (NEW β€” Prevents Repeat Work)

Every major project now requires a status file. This is the single source of truth for "where are we?"

memory/peg/ALEX-PEG-STATUS.md β”‚ β”œβ”€β”€ βœ… Completed (DON'T REDO) ← Things that are DONE. Don't touch. β”‚ └── Pricing engine β€” penny perfect βœ… β”‚ └── Logo rendering β€” hardcoded βœ… β”‚ β”œβ”€β”€ πŸ”„ Current State ← What's working RIGHT NOW β”‚ └── Drawing system: production ready β”‚ β”œβ”€β”€ ⏭️ Next Steps ← What to do NEXT (in order) β”‚ └── 1. Factory drawing template β”‚ └── πŸ“‹ Solved Problems ← Reference table of past fixes └── Logo issue β†’ root cause β†’ fix β†’ date
Rule for Kevin: If an agent seems to be redoing work, say: "Check your status file before starting." This forces them to read what's already been done.

memory_search β€” The Recall System

Agents have a memory_search tool that does semantic search across all memory files. Before answering questions about past work, they're supposed to search first. But it's not perfect β€” short status files with clear labels work better than long rambling notes.

How to Help Your Agents Remember

  • End sessions explicitly: "OK, save your progress and update your status file" forces a checkpoint
  • Ask "what's your status on X?" before assigning related work β€” makes them read the file
  • Keep MEMORY.md clean: Only high-level facts. Details go in project files.
  • Use "Don't redo" lists: Explicitly listing solved problems prevents re-investigation

πŸ”„ Session Lifecycle

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ WAKE UP │───►│ LOAD │───►│ WORK │───►│ SAVE β”‚ β”‚ (blank) β”‚ β”‚ FILES β”‚ β”‚ SESSION β”‚ β”‚ PROGRESS β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚ β”‚ Read MEMORY.md Update STATUS.md Read AGENTS.md Update memory/date.md Read context files Update MEMORY.md if needed β”‚ β”Œβ”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β” β”‚ SESSION β”‚ β”‚ ENDS β”‚ β”‚ (amnesia) β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

The critical step most agents skip is the SAVE step. That's what we just fixed in Alex's AGENTS.md β€” mandatory progress saves.

πŸ’Ό Valentina β€” Executive Operations

Model
Claude Opus 4 ($15/$75 per MTok)
Machine
Mac Mini M4 Pro 64GB
Tailscale
100.87.49.18
Primary Channel
iMessage (+13109242990)

Responsibilities

  • Morning briefings (5 AM daily) β€” email, calendar, portfolio, weather
  • Email classification & triage (every 30m M-F)
  • V-PEG pricing engine (penny-perfect across 35 line items)
  • Zoho CRM task creation & management
  • QuickBooks financial briefings (7 AM M-F)
  • Team coordination β€” can SSH to Alex and Sage
  • Customer pricing portal & training hub maintenance
  • Zoho Cliq monitoring for team messages

Key Files

memory/peg/VALENTINA-PEG-STATUS.mdV-PEG pricing engine status
memory/email-classification-log.jsonlEmail processing logs
memory/heartbeat-state.jsonLast check timestamps
valentina_peg/V-PEG source code

Cron Jobs

JobScheduleWhat
Morning Briefing5:00 AM ET dailyEmail + calendar + portfolio + weather
Financial Briefing7:00 AM M-FQuickBooks data summary
Email ClassificationEvery 30m M-FClassify & triage inbox
Afternoon Digest2:00 PM M-FEmail summary
GA Weekly ReportMon 6 AMGoogle Analytics summary
Password RotationSun 8 PMCustomer pricing portal password

πŸ”§ Alex β€” Technical Engineering

Model
Claude Opus 4 ($15/$75 per MTok)
Machine
Mac Mini M4 Pro 24GB
Tailscale
100.81.25.123
Primary Channel
iMessage

Responsibilities

  • V-PEG drawing generation (LightPanel, LightPaper, Media Frames)
  • PEG system reverse-engineering & documentation
  • Technical integrations and code development
  • Engineering documentation

Key Files

memory/peg/ALEX-PEG-STATUS.md⭐ Master PEG project status (NEW)
memory/peg/PEG-SUMMARY.mdPEG system overview
memory/peg/PEG-ARCHITECTURE.mdTechnical architecture
valentina_peg/Drawing generator code
Known Issue β€” Repeat Work: Alex has a pattern of re-solving problems he's already fixed (logo rendering was fixed 3+ times). The new ALEX-PEG-STATUS.md with "Solved Problems" table and mandatory progress saves should prevent this going forward.

🎨 Sage β€” Marketing & Content

Model
Claude Opus 4
Machine
TBD
Tailscale
100.99.253.24
Status
⚠️ SSH unreachable (Feb 25)

Planned Responsibilities

  • Social media content creation (LinkedIn, Instagram, Facebook, TikTok)
  • Competitive analysis & trend monitoring
  • Website content & SEO optimization
  • Blog writing & thought leadership
  • Content calendar management
  • Sales follow-up email campaigns

πŸ’¬ How to Talk to Your Agents

Best Practices

  • Be specific about what you want: "Update the pricing for Basic LightPanel" > "Fix the pricing"
  • Reference project names: "Check V-PEG status" tells the agent exactly which status file to read
  • End work sessions explicitly: "Save your progress" triggers a status file update
  • Ask for status before assigning: "What's your status on PEG drawings?" forces a file read

Useful Commands

You SayWhat Happens
"What's your status on [project]?"Agent reads the project status file and gives you a summary
"Save your progress"Agent updates status files and memory
"Check your status file before starting"Prevents redoing completed work
"Remember this: [fact]"Agent writes to memory file
"What do you know about [topic]?"Agent runs memory_search across all files
/statusShows session info (model, cost, uptime)

Cross-Agent Communication

You ──iMessage──► Valentina ──SSH──► Alex β”‚ └──────SSH──► Sage Valentina coordinates. She can: β€’ SSH into Alex/Sage to check their files β€’ Read their status files β€’ Update their memory files β€’ Run commands on their machines
Pro Tip: If you need Alex to do something, you can either message Alex directly OR tell Valentina to coordinate. Valentina can check Alex's status files, see what he's been working on, and even set up tasks.

When Topics Switch Mid-Conversation

This is the main source of "lost context." When you switch topics in the same session:

  • The agent's conversation window fills up with the new topic
  • Old topic context scrolls out of the active window
  • If progress wasn't saved before switching, it can be lost

Solution: Before switching topics, say "Save progress on [current topic], then let's talk about [new topic]."

πŸ“± Sharing Work Between Claude.ai and Agents

You often do deep work in Claude.ai (web), then want your agents to continue it. Here's how:

Best Methods

MethodHowBest For
Google DocPaste Claude output β†’ save to Drive β†’ agent reads via gogLong analyses, documents
Direct pasteCopy key parts into iMessageQuick handoffs, short content
Workspace fileSave as .md file in agent's workspaceCode, data, reference material
Claude.ai shared links DON'T work β€” they render client-side with JavaScript. Agents see "Loading..." when they try to fetch them. Always copy the content instead.

πŸ“Š Active Projects

V-PEG (AI Quoting Engine)

Lead: Alex (drawings) + Valentina (pricing)

ComponentStatusOwner
Pricing EnginePENNY PERFECTBoth (verified)
LightPanel DrawingsPRODUCTION READYAlex
LightPaper DrawingsPRODUCTION READYAlex
Intake PipelineBUILTValentina
Zoho CRM SyncBUILTBoth
Factory DrawingsNOT STARTEDAlex
Custom ShapesNOT STARTEDTBD

EOS Command (Email Automation)

Lead: Valentina

ComponentStatusNotes
Email ClassificationLIVEEvery 30m M-F
David Ross Templates3 TEMPLATESGeneral, Quote, Troubleshooting
Response AccountabilityLIVEAuto-creates Zoho tasks
Auto-Replies (Phase 2)PLANNEDAfter Phase 1 validated
Agent NetworkPLANNEDDonna Router + specialist agents

Customer-Facing Portals

PortalURLStatus
Customer Pricingpricing-customer.eoslight.comLIVE
Training Hub (Phoenix)phoenix.eoslight.comLIVE

Infrastructure

SystemStatusNotes
Valentina SSH β†’ AlexWORKINGalexchen@100.81.25.123
Valentina SSH β†’ SageDOWNConnection refused (Feb 25)
Tailscale NetworkCONNECTEDAll machines on Tailscale
Zoho CRM APICONNECTEDVia Maton gateway
QuickBooks APICONNECTEDVia Maton gateway
Google AnalyticsCONNECTEDWeekly reports running

πŸ”§ Common Issues & Fixes

Agent seems to have forgotten everything

Cause: New session started, MEMORY.md might be outdated or the specific project file wasn't read.
Fix: Say "Read your status file for [project] and tell me where you left off."

Agent is redoing completed work

Cause: Status file doesn't have the task in "Completed" or agent didn't read it.
Fix: Say "Check your status file before starting. Look at the Completed section."

Agent can't connect to a service

Cause: API token expired, service down, or auth needs refresh.
Fix: Check the service directly, then update TOOLS.md or memory files with new credentials.

Can't SSH to another agent

Cause: Machine off, SSH disabled, or Tailscale disconnected.
Fix: Check Tailscale admin console. If machine is on but SSH fails, Remote Login may need re-enabling in System Settings β†’ General β†’ Sharing.

Agent gateway is down

Cause: Process crashed or machine restarted.
Fix: SSH in and run openclaw gateway start --force, or use the watchdog script.

Claude.ai shared links don't work

Cause: Shared links render client-side. Agents can't extract the content.
Fix: Copy/paste content into iMessage, a Google Doc, or a workspace file instead.

πŸ”‘ Quick Reference β€” Key IPs & Access

AgentTailscale IPSSH UserSSH Command
Valentina100.87.49.18valentinavossssh valentinavoss@100.87.49.18
Alex100.81.25.123alexchenssh alexchen@100.81.25.123
Sage100.99.253.24TBDCurrently unreachable