company-manager-agent

Executive-facing company operations assistant built on OpenClaw, Telegram, SQLAlchemy, and SQLite.

This project is a proof of concept for a CEO or senior operator who wants to ask natural-language questions like:

• "Give me a company snapshot and highlight the biggest delivery risk."
• "What meetings do I have this week?"
• "Assign a high-priority task to the operations lead."
• "Move CEO Ops Review to next Monday."

The assistant answers from a local company database, writes operational changes back to that database, and is exposed through OpenClaw so it can be used from Telegram.

What this project demonstrates

• End-to-end executive chat workflow from Telegram to OpenClaw to a live SQLite company database
• Read operations for company health, employees, products, orders, meetings, and tasks
• Write operations for task creation, task status updates, meeting creation, meeting rescheduling, meeting cancellation, and meeting deletion
• A custom OpenClaw skill that teaches the agent how to answer operational questions from the local company dataset
• A deterministic SQLAlchemy-backed data layer that can be seeded for repeatable demos

Architecture

Telegram
  -> OpenClaw Gateway
    -> company-manager skill
      -> Python CLI
        -> SQLAlchemy
          -> SQLite

Core pieces:

• company_manager/models.py Defines the data model for companies, departments, employees, products, customers, orders, meetings, meeting attendees, and tasks.
• company_manager/tooling.py Implements the query and mutation layer used by the agent-facing CLI.
• scripts/company_manager_tool.py Exposes operational read and write commands in a simple CLI that returns structured JSON.
• scripts/migrate.py Creates the initial schema and applies lightweight schema updates needed for local evolution.
• scripts/seed_demo_data.py Seeds a demo company, Northstar Dynamics, with realistic sample operating data.
• openclaw_skills/company-manager/SKILL.md Teaches OpenClaw when and how to use the local company manager tooling.

Current capabilities

Read visibility

• Company summary: employee count, product count, order count, open orders, active tasks, booked revenue, task breakdown
• Employees: browse people by department, title, email, and location
• Products: catalog, SKU, category, price, inventory
• Orders: recent orders, filtered orders by status, line items, customer, order totals
• Meetings: upcoming or full meeting schedule, attendees, agendas, status
• Tasks: task owner, priority, status, due dates, descriptions

Write actions

• Create tasks
• Update task status
• Schedule meetings
• Reschedule meetings
• Cancel meetings
• Delete meetings
• Clean up duplicate operational meetings directly from the assistant workflow

Demo dataset

The seeded database contains one demo company:

• Northstar Dynamics

The sample data includes:

• 5 departments
• 23 employees
• 6 products
• 18 customers
• 36 orders
• multiple executive and operating meetings
• live tasks across sales, engineering, operations, and customer success

Database location:

• data/company_manager.db

Setup

cd /home/shreyak/programming/company-manager-agent
python3 -m venv .venv
.venv/bin/pip install -e .
.venv/bin/python scripts/migrate.py
.venv/bin/python scripts/seed_demo_data.py

Local CLI usage

Read commands

.venv/bin/python scripts/company_manager_tool.py summary
.venv/bin/python scripts/company_manager_tool.py employees --department Engineering --limit 10
.venv/bin/python scripts/company_manager_tool.py products --limit 10
.venv/bin/python scripts/company_manager_tool.py orders --status pending --limit 10
.venv/bin/python scripts/company_manager_tool.py meetings --limit 10
.venv/bin/python scripts/company_manager_tool.py meetings --all --include-cancelled --limit 20
.venv/bin/python scripts/company_manager_tool.py tasks --limit 10

Write commands

.venv/bin/python scripts/company_manager_tool.py create-task \
  --title "Prepare weekly board summary" \
  --description "Summarize revenue, delivery risk, and hiring bottlenecks." \
  --owner-email ethan.johnson40@northstar.example \
  --priority high \
  --due-date 2026-04-12

.venv/bin/python scripts/company_manager_tool.py update-task-status \
  --task-id 5 \
  --status in_progress

.venv/bin/python scripts/company_manager_tool.py schedule-meeting \
  --title "CEO Ops Review" \
  --start-at 2026-04-13T15:00:00 \
  --duration-minutes 45 \
  --location Zoom \
  --agenda "Review shipping backlog and assigned actions." \
  --attendee-email ethan.johnson40@northstar.example \
  --created-by agent

.venv/bin/python scripts/company_manager_tool.py reschedule-meeting \
  --meeting-id 8 \
  --start-at 2026-04-14T15:00:00

.venv/bin/python scripts/company_manager_tool.py cancel-meeting --meeting-id 8
.venv/bin/python scripts/company_manager_tool.py delete-meeting --meeting-id 8

OpenClaw integration

The OpenClaw skill for this project lives here:

• openclaw_skills/company-manager/SKILL.md

OpenClaw needs that folder added to skills.load.extraDirs in ~/.openclaw/openclaw.json.

Example:

{
  "skills": {
    "load": {
      "extraDirs": [
        "/home/shreyak/programming/company-manager-agent/openclaw_skills"
      ]
    }
  }
}

You can verify discovery with:

openclaw skills info company-manager

Telegram workflow

Once OpenClaw is running and Telegram is configured, the assistant can be used from chat as a company manager bot.

Example prompts:

• "Give me a company snapshot and highlight the biggest delivery risk."
• "What meetings do I have this week?"
• "Assign a high-priority task to Ethan Johnson to reduce shipment delays."
• "Move CEO Ops Review to next Monday."
• "Delete the earlier duplicate CEO Ops Review meeting."
• "Show me pending orders and the largest backlog risk."

Example operating loop

1. Executive asks a question in Telegram.
2. OpenClaw receives the message and routes it to the main agent.
3. The company-manager skill instructs the agent to use the local Python CLI.
4. The CLI reads or writes the SQLite database through SQLAlchemy.
5. The assistant returns a human-readable answer or confirmation back to Telegram.

This gives you a narrow but real closed loop:

• conversational input
• operational database reads
• operational database writes
• human-readable output in chat

Project status

This is a working proof of concept, not a finished production system.

What is solid today:

• local data model
• repeatable seeding
• read and write operations
• OpenClaw skill loading
• Telegram-facing demo workflow

What is intentionally simplified:

• SQLite instead of Postgres
• one demo company instead of multi-tenant isolation
• no auth or RBAC around business actions
• no audit UI beyond DB state and OpenClaw session logs
• no external ERP, CRM, HRIS, or calendar sync yet

Recommended next steps

• Move from SQLite to Postgres
• Add explicit action audit logs
• Add richer analytics queries for revenue, delays, and operating KPIs
• Add role-aware permissions for CEO, COO, VP, and manager actions
• Add connectors for real calendar, CRM, ERP, and ticketing systems
• Expose direct command-dispatch for common scheduling flows to reduce LLM ambiguity further

License

MIT