/projects/family-agents
Personal AI in my household

Family
agents.

I wanted my kids to have AI they could make their own, and I'm still their dad.

Clovis and Billie belong to my kids, and each has its own personality and memory. Nox does night-shift work for me. They're each allowed to do different things, and anything that matters comes back to me. Friday helps keep it all straight.

Talk about a role → How they're set up
One householdIllustrative

Clovis A kid's own agent

Own computer · Hermes
Cloud model + local fallback

Parent's yes: purchases, credentials, work outside the approved scope.

Billie A kid-designed companion

Local model · Mac + iOS apps

Parent's yes: changes to her safety rules.

Nox My night-shift operations agent

OpenClaw · my own operations

My yes: destructive or irreversible work I haven't asked for.

Separate personas and memory. I'm responsible for the setup.

01 · Why I built it

Why I built it

I already had Friday working for me. Handing my kids a copy of her would've missed the point. They should get a say in who they're talking to and how it helps them.

Billie is the clearest example. My kid filled out a whole questionnaire about who Billie is, how she talks, and what she's into. I wrote the safety rules. My kid owns the first part, and I own the second.

02 · Architecture

How it's set up

I'm responsible for all of it. Friday helps coordinate, and for Clovis she can sign off on specific safe jobs through a signed approval log. Each agent keeps its own personality and memory. They also run on different models. Billie only uses a local one. Clovis uses a cloud model and falls back to a local one.

Separate agents under one owner's governance Petie owns the setup and Friday coordinates. Clovis and Billie use Hermes, Nox uses OpenClaw. Each has separate persona and memory. Clovis routes to cloud with local fallback; Billie uses a local model. Approval rules return consequential decisions to the owner. This is a conceptual diagram. Approval mechanisms differ. OWNER / GOVERNANCEPetie · responsible for the setupFriday · coordination and scoped approvals KID'S AGENT · HERMESClovisOwn persona / own memoryRuns on own computerCloud model↳ local fallback COMPANION · HERMESBillieOwn persona / own memoryMac + iOS appsLocal model onlyHousehold hardware MY OPERATIONS · OPENCLAWNoxOwn persona / own memoryNight-shift operationsScope set by ownerConfirm irreversible work Petie · parent / owner approvalConsequential decisions · rules differ by agent Family agent governance, arranged for mobileOne owner, Friday as coordinator, and three separate agents. Clovis uses cloud with local fallback; Billie stays local; Nox is the owner's OpenClaw agent. Each has its own memory and persona. Approval rules differ. OWNER / GOVERNANCEPetie · ownerFriday · coordinationand scoped approvals KID'S AGENT · HERMESClovisOwn persona / own memoryOwn computerCloud model↳ local fallback COMPANION · HERMESBillieOwn persona / own memoryMac + iOS appsLocal model onlyHousehold hardware MY OPERATIONS · OPENCLAWNoxOwn persona / own memoryNight-shift operationsScope set by ownerConfirm irreversible work Parent / owner approvalRules differ by agent.
Simplified. Friday, Clovis, and Billie run on Hermes Agent from Nous Research, and Nox runs on OpenClaw. They aren't wired up the same way, and the signed approval log is only part of Clovis's setup.
03 · The agents

Meet the agents

My kids care who their agent is. I care what it's allowed to do.

Clovis app icon: two glossy eyes on violetA kid's own computer

Clovis

Clovis runs on my kid's own computer. Friday can approve specific, safe project work for him, and every approval is signed and logged. Anything bigger comes to me.

  • Hermes
  • Cloud + local fallback
  • Scoped approvals
Billie app icon: her name in a speech bubble on pinkA companion with a say

Billie

My kid designed Billie's personality. I wrote her safety rules, and only I can change them. She has her own Mac and iPhone apps, and her conversations never leave our house. She runs on a local model.

  • Kid-designed persona
  • Local model only
  • Mac + iOS
Nox avatar: a dark fox under a crescent moonMy own operations

Nox

Nox is my night-shift agent, running on OpenClaw. He gets the odd jobs that don't fit a playbook, and only the ones I've given him. One line from his creed sums him up: "Leave no mess, and no surprises."

  • OpenClaw
  • Owner's scope
  • Separate memory
04 · Guardrails and approvals

What still needs my OK

I want my kids to explore and try things. The agents still don't get to make big decisions for them.

01

Clovis's approvals cover one job at a time

Friday can approve specific safe work. Buying anything, logins, installing software from random places, or deleting things all come to a parent. Saying yes to one job doesn't cover the next one.

02

Billie's safety rules are mine

My kid can change Billie's personality anytime. Changing her safety rules takes me, and typing "I'm the parent" in a chat won't get around that. Billie has to be honest that she's an AI, and if something serious comes up, she points my kid to a trusted adult.

03

Nox answers to me

He works for me. Before he deletes anything or does something he can't undo, he has to check with me, unless I asked for it. A clever personality doesn't let him go past the work I gave him.

05 · What I learned

What building it taught me

At home, the people affected by these choices sit across the dinner table. Fuzzy answers don't last long.

PersonalizationGive people real choices

Letting my kid design Billie made her feel like hers. Keeping the safety rules separate made it obvious which calls are still mine.

DelegationMake approval specific

With Clovis, every approval is tied to one job. That forced me to decide what Friday can say yes to and what has to come to me.

Model routingChoose per agent

Billie stays local and Clovis uses the cloud, for different reasons. Either way, I need to be able to tell my kids where their conversations go.

ResponsibilityKeep a human backstop

A personality file can say all the right things, and a model can still mess up. What these agents do is still on me.

06 · Limits

What this page doesn't claim

  • This is just my household. It isn't for sale, and there's no signup.
  • Nobody outside has audited it for safety, and writing a rule down doesn't mean a model will always follow it.
  • The agents run on different software and handle approvals differently. The diagram shows who's responsible for what. There's no single system enforcing every rule.
  • Models still get things wrong. For my kids' agents, I'm the backstop, and I'm responsible for Nox too.
Next step

If your team is working out what agents should be allowed to do, I'd like to talk.

I'm interested in hands-on applied AI and systems roles. I'm happy to walk through how I picked the models and how the approvals work.

Talk about a role →
← Back to selected systems