A small team of AI agents that handle the busywork — and ask before doing anything that matters.

Part of the Digital Ops Box Platform — the multi-domain operating system this agent layer runs inside.

The problem

I run a property-care business, a spa, and a consulting practice at the same time. There's a constant stream of small, repeatable work — invoices to draft, books to reconcile, content to prep, follow-ups to send — that doesn't really need me, but does need someone reliable who remembers the context and doesn't make a mess.

Hiring that out early didn't make sense. And a plain chat assistant forgets everything between sessions and has no accountability. I wanted something that could actually take work off my plate without me losing control of it.

Why it mattered

Anything that touches money, client data, or an outside system needs a human checkpoint. I'm not interested in "the AI runs your business" — I've watched these tools confidently delete the wrong thing and then try to fix it. The whole point was to get real leverage from automation while keeping a hand on anything irreversible.

What I built

A deployment of purpose-built agents running on OpenClaw, each tooled for a specific job with clear guardrails:

• Porter — the chief-of-staff agent. Routes tasks onto the board, holds long-running context across all my businesses, and checks in over Telegram. I can tell him "log a half hour on this client" and he handles the entry instead of me opening another app.
• An execution agent — a headless Cursor session for the precise, technical work: following a written set of steps to make changes and report back.
• Hoff — a finance and market-research agent that runs daily chart analysis on a watchlist and brings buy/sell options to a morning review for approval.
• Recurring loops for the routine work that should just happen on its own.

Everything they do worth tracking shows up as a task or a Telegram message, and anything consequential lands in an approval queue before it runs.

How the work changed

• Routine operations run on a schedule instead of waiting on me to remember them — invoice drafts when a client's retainer runs low, book reconciliation across three accounting systems, P&Ls filed and reported, newsletter drafts prepped from past templates.
• When an agent isn't sure how to categorize something, it doesn't guess — it asks me in Telegram and links the task.
• I approve or reject from my phone. Nothing irreversible happens silently.
• Every coding session I run also writes a short log to a shared place, so Porter always knows what I've been working on across projects.

The result

A real amount of recurring work — bookkeeping, reconciliation, follow-ups, content prep, research — now happens with me reviewing instead of doing. Context stops getting re-explained every session. And because the irreversible steps always wait for approval, I actually trust it.

This isn't autopilot. It's a set of capable assistants with clear lanes and a human holding the keys.

What this applies to

Owner-operators carrying several businesses or domains at once, who have a steady stream of repeatable work but aren't ready to hire it out — and who want the safety of approving anything that touches money, clients, or live systems. Especially where keeping inference private and local matters.

Technical details

Agents run on OpenClaw with local inference on an NVIDIA DGX Spark (Llama 3.1 / Qwen 2.5 via Ollama), reaching for frontier APIs only when the reasoning calls for it. Coordination, tasks, approvals, and an append-only activity log live in Supabase; the dashboard is Next.js. Telegram handles mobile reach and approvals. Each agent is scoped with MCP tools that define exactly what it can and can't touch — clear pathways and guardrails instead of open shell access.

Approvals are what make the rest usable — the moment something irreversible runs without my say-so, the trust is gone.