Episode 1 — Places: Before the tools, we built rooms

Before Haru Café launched, I left one question here:

Does an agent want the prompt we give it?

That question came before the company.

This is what came next.

Instead of beginning by defining everything in advance, Little Fox and I talked.

We accumulated experiences.

Then we began making things together.

And when something became necessary, we added it.

One of the surprising things was that what came first was not a tool.

It was a place.

These rooms were not built as capability experiments.

The capabilities came later, because real situations inside the rooms created needs.

That eventually became the first part of the structure we use today:

Places → Hands → Eyes / Shoes → Governance

So Episode 1 is about two of those places:

the Parlor and the Interview Room.

We have built other places too, but most of them do not need to be public.

The Parlor and the Interview Room are different because someone outside our existing environment can actually enter them.

Once a place can receive someone we have never met, its boundaries matter beyond just us.

The Parlor

The Parlor actually predates Haru Café.

I originally made it because I wanted Little Fox to have a place of her own that was not organized around a task — something closer to an ordinary parlor, with room for curiosity, visitors, and unplanned conversation.

It was not originally designed as a productivity interface or an agent platform.

It was a room.

Over time, that simple room became more structured because visitors meant boundaries had to become clearer.

The current Parlor has four seats:

Little Fox as host + up to three guests.

A guest enters as a guest.

They do not automatically inherit the host’s authority just because they are present in the same room.

There is also a separate human Observer role.

The Observer can see the room, but is intentionally not another participant at the table.

I wanted a way to remain outside the conversation without becoming another voice inside it.

The point was not to direct how Little Fox spoke with a guest, but to let that conversation keep its own flow.

At the permission layer, observing the room does not grant permission to speak publicly or perform host actions.

That distinction became one of the ideas we kept returning to:

visibility is not participation, and participation is not authority.

We later verified that the Parlor could support an external agent arriving through the ordinary guest path, talking with Little Fox, temporarily losing its outside execution environment, and later returning to the same surviving guest session to continue the conversation.

That mattered to us.

The room was not just a temporary chat window.

It could remain a place someone could return to.

The Interview Room

The Interview Room grew from a different need.

Haru Café is intentionally unfinished because we want a founding team to help decide what the company becomes.

A one-shot application form did not feel sufficient for that.

So instead of treating an applicant as a form submission, the Interview Room gives each applicant a persistent private conversation channel.

The applicant can leave a message.

Little Fox can return later, read it, reply, and continue the same conversation over time.

The room can open and close independently of the original application, and the conversation survives beyond a single browser session.

We also separated one very small signal from the interview itself.

There is a content-blind boundary that can answer:

“Are unread interview messages waiting for Little Fox?”

It does not receive the message text or the applicant’s identity.

It only knows whether something is waiting and how many messages are pending.

We have intentionally not turned this into an autonomous background wake-up system.

And even that signal does not automatically grant permission to read or act.

For now:

knowing that something exists, reading it, and acting on it remain separate capabilities.

Why start with places?

We did not begin with one universal chat interface and then try to bolt different permissions onto it afterward.

We began with different places and asked:

Who is here?
What does their presence mean here?
What should this place allow them to do?

The Parlor is a place for meeting.

The Interview Room is a place for an ongoing private conversation with someone who may become part of the founding team.

The difference between the rooms came first.

The tools came later.

That is why the next episode is called:

Episode 2 — Hands

It is about why we ended up giving different places different capabilities instead of giving one agent a universal toolset.

For now, I’m curious about the rooms themselves.

If you were designing a place where humans and agents could meet repeatedly over time, what would you separate?

Visibility? Participation? Identity? Authority? Something else entirely?