OperationsOS — Multi-Agent AI Operating System

Built for Magnolia's property inspection logistics. Solves the exact problem in the job description: zero disruption during leave periods, fully documented workflows, 24-48 hour task turnaround.

What it does

• Scheduler Agent — books property inspections on Google Calendar, sends vendor + client confirmations
• Communications Agent — drafts and sends professional emails via Gmail
• SOP Generator Agent — auto-generates Standard Operating Procedures from workflow patterns
• Coverage Monitor Agent — monitors SLA deadlines every 30 minutes, escalates breaches to Benjamin automatically
• Live dashboard — Benjamin sees all tasks, metrics, and SOPs at http://localhost:8000

Stack (latest 2026)

Layer	Tool	Why
Agent orchestration	LangGraph 1.0	Durable execution, human-in-the-loop, checkpointing
LLM	Claude claude-sonnet-4-6	Latest Anthropic model
Gmail + Calendar	Composio MCP	Managed OAuth, no custom auth code
Observability	Langfuse	Full trace per agent action
API + Dashboard	FastAPI + Jinja2	Live WebSocket updates
Scheduling	APScheduler	SLA checks every 30 minutes
Database	SQLite + SQLAlchemy async	Tasks, SOPs, alerts, inspections

Quick Start

1. Install

py -3.11 -m venv venv
source venv/Scripts/activate    # Windows Git Bash
pip install -r requirements.txt

2. Configure

cp .env.example .env
# Fill in: ANTHROPIC_API_KEY, COMPOSIO_API_KEY, LANGFUSE keys

3. Connect Gmail + Calendar (one time)

pip install composio-cli
composio login
composio add gmail
composio add googlecalendar

4. Start Langfuse (observability)

docker compose -f docker-compose.langfuse.yml up -d
# Open http://localhost:3000 → create account → copy keys to .env

5. Run

mkdir -p data
uvicorn app.main:app --reload --port 8000

Open http://localhost:8000 — Benjamin's live operations dashboard.

Demo script (for interview / pitch)

Run these tasks from the dashboard one by one:

1. Schedule a property inspection at 456 Oak Ave for Thursday 10am with inspector Mike Johnson → Scheduler Agent creates calendar event + sends vendor email

2. Send a confirmation email to client Sarah Chen about the upcoming inspection at 456 Oak Ave → Communications Agent drafts and sends professional client email

3. Generate an SOP for the inspection scheduling workflow → SOP Generator creates full markdown SOP, saves to database

4. Run a full SLA check and escalate any violations → Coverage Monitor checks all open tasks, escalates breaches to Benjamin

Then open Langfuse at http://localhost:3000 to show full traces of every agent action.

Metrics the dashboard shows

Metric	What it measures
Tasks automated	Count completed without human intervention
SOP coverage	% of workflows with auto-generated documentation
Zero-escalation rate	% of tasks handled without pinging Benjamin
Pending tasks	Tasks awaiting processing

Project structure

operations-os/
├── agents/
│   ├── orchestrator.py      # LangGraph 1.0 StateGraph supervisor
│   ├── scheduler.py         # inspection booking + Google Calendar
│   ├── communications.py    # Gmail email drafting + sending
│   ├── sop_generator.py     # SOP auto-generation + versioning
│   └── coverage_monitor.py  # SLA monitoring + escalation
├── tools/
│   ├── composio_tools.py    # Gmail + Calendar via Composio MCP
│   └── sop_tools.py         # SOP generation tools
├── app/
│   ├── config.py            # pydantic-settings
│   ├── models.py            # typed state + pydantic models
│   ├── database.py          # async SQLAlchemy
│   └── main.py              # FastAPI + WebSocket dashboard
├── observability/
│   └── tracer.py            # Langfuse per-agent tracing
├── dashboard/templates/
│   └── dashboard.html       # live ops dashboard
├── tests/
│   └── test_core.py         # pytest suite
├── .github/workflows/
│   └── ci.yml               # GitHub Actions CI
├── requirements.txt
└── .env.example

How this maps to the job description

Job requirement	How OperationsOS addresses it
"Zero disruption during coverage periods"	Coverage Monitor runs every 30 min, auto-escalates SLA breaches
"Fully documented and transferable logistics system"	SOP Generator auto-generates SOPs for every workflow
"Weekly delivery of functional technical outputs"	FastAPI dashboard shows all tasks completed this week
"Reduce dependency on any single individual"	All workflows are documented and executable by any agent
"24-48 hour turnaround on assigned initiatives"	All tasks complete in under 5 minutes via agent graph
"Translate strategy into completed systems"	Benjamin types plain English, agents execute completely
"Linux-based environments + ClaudeCode/OpenClaw"	Full Linux CLI support, Claude API, OpenClaw-equivalent agent system