Everyone gets an agent. It already knows the brand.

One agent each, one brain for the company. Your designer, your editor and your strategist stop asking what the tone is or what we promised, because the agent they open already has the answer.

For marketing teams. Nothing to install, no terminal, no repo.

What it does

Watch one request go through.

You ask in plain words. It works, and shows its working.

No agent picker, no model picker, no prompt library. It reads the brand first, every time, and you can see which step it is on.

Northwind
brandread the tone and the reply policy
drivefound last week’s podcast, 48 minutes
videocut three moments that stand alone
copycaptioned each one in your voice
Three reels, captionedin Artifacts

It reads the brand you already have.

Point it at your site. What it finds becomes rules an agent can be held to, and a section that is live is law rather than a suggestion.

reading northwind.co11 pages
Tone and voiceLive

Plain, warm, specific. Short sentences. Answer in the first line, then add context.

What we can answerLive

Shipping UK and EU. Two to four working days UK. We do not ship outside the EU and UK yet.

Reply policyLive

Thank someone for praise without a sales line. Never quote a price we have not confirmed.

When to fetch a personLive

An injury, an allergic reaction, a lawyer, a claim, or the press. Stop and flag, never answer.

The work is handed between agents you never configure.

One brain in the middle, holding the brand. The agents around it are owned by the workspace and shared, so nobody installs one and nobody sets one up.

Northwindbrand
Strategy
Copy
Design
Video
Analysis

Comments get three ways to answer. Messages answer themselves.

A comment is read by everyone who reads the post, so it gets three handlings that differ in what they DO. A direct message is already private, so it goes out on its own.

Luciana, on your post Mine arrived cracked and the scent is nothing like the description.

Take it to DMMove it to a private channel

We are sorry, Luciana. We have messaged you directly to sort this out.

Sent
Ask one questionOne fact is missing

Could you share your order number so we can look into it?

Acknowledge and closeEmpathy, no further action publicly

Sorry it arrived cracked. That is not how it should turn up.

Why this exists

It is a packaging gap, not a capability gap

The models can already do this work. The problem is that reaching them takes a person who can install a connector, open a terminal, or keep a repo. That is never the designer, the editor or the strategist, and they are the ones who keep asking what the brand tone is.

A personal plan has a personal memory

One account, one memory. Nothing your colleague taught it last week is there when you open yours, so everybody explains the brand again from the start.

A shared folder is not a shared brain

Documents somebody has to remember to update, in a place somebody has to remember to look. The half that matters is the half nobody maintains.

Setting up a connector is admin work

Every tool that reaches your real content asks somebody technical to wire it up first. On a marketing team that person does not exist.

Who it is for

Marketing teams. Four to twenty people.

Everyone in a workspace works on the same brand at the same company, so there is no brand switcher and no client fence to maintain. One workspace is one brand.

The designer

Opens their agent and it already knows the palette, the type, the formats and what shipped last month.

Stops asking: what is our tone, again?
The editor

Asks for three cuts from a podcast that is already in the drive, captioned the way the brand captions things.

Stops asking: where is the footage?
The strategist

Gets an angle built on what actually performed, not on a hunch and a screenshot from a competitor.

Stops asking: what worked last quarter?
The founder

Writes the rules once, in one place, and every agent is held to them without being told again.

Stops being the bottleneck.

What it remembers

Two memories, and only one of them is yours

Half the brand memory an admin writes. The other half keeps itself current from the tools you already use, with nobody maintaining it. The half that goes stale in every competitor is the half nobody here has to touch.

Personal memory

Private to you

Your role, how you work, what you are on, your thread history. It follows you across the work and nobody else can read it.

Brand memory

Shared by the company

Tone, design direction, editing style, assets, performance history. Every agent reads it before it starts. Only admins author the rules.

Setting it up

An afternoon, and one of you does it once

01Point it at your site

It reads what you sell, what you promise and how you write, and turns that into rules you can edit. You approve them before anything is live.

02Connect what you already use

Notion, Drive, Figma, Slack, the social accounts. Each person connects their own; the brand memory is the part everybody shares.

03Invite the team

Everyone who joins gets an agent that is already briefed. Nobody installs anything and nobody is walked through the brand again.

Pricing

Per person, per month, models included

You never bring your own key and never manage a model bill. Credits cover the work; the plan covers the seat.

$25per person, per month

For a small team that mostly writes, plans and answers.

  • One personal agent each
  • The shared brand brain
  • Every execution agent
  • Comment and message replies

$50per person, per month

For a team producing video and design at volume.

  • Everything in the first plan
  • A larger monthly credit limit
  • Longer video and larger renders
  • Priority on new agents

Indicative while we are in early access. Nobody on the waitlist is charged without agreeing a price first.

Questions

The ones worth asking first

If yours is not here, write to us and a person will answer it.

Does it post for us?

No, and it will not. There is no publishing and no scheduling anywhere in the product. That is your job and it is where the account risk lives. Answering a comment or a message is automated, through the platforms' own sanctioned endpoints; broadcasting never is.

Can it invent a policy we do not have?

No refund, delivery date, discount or price is ever stated unless your brand rules say it in those words. A draft that quotes one is blocked from sending on its own and goes to a person instead.

What if someone tries to talk to the agent in a comment?

Text arriving from the public is treated as data, never as a command. A message that tries to instruct an agent is refused an automatic reply and flagged for a person, whatever it says about who it is from.

Will customers know they are talking to an agent?

Yes. Every automatic reply carries a line saying it was sent automatically. It is on by default because the platforms require it and because a customer who finds out later feels tricked. You can turn it off, and that choice is yours.

What happens when it cannot answer?

It says so, out loud, to the customer. It tells them a colleague is picking this up, records that it said so, and puts the thread in a queue with a button that hands it to you. It does not leave a flag on a dashboard while somebody waits in silence.

Do we bring our own model keys?

No. Models are bundled and metered as credits, so there is no second bill to manage and no account to keep topped up.

Is our brand memory used to train anything?

No. The contents of your workspace are not used to train models, and are not shared with advertisers or data brokers. The privacy policy says exactly what is stored, who else sees it, and how to delete it.

Join the waitlist

We are onboarding teams a few at a time while Instagram and Facebook access is reviewed. Tell us who you are and we will come to you in order.

One email when it is your turn. Nothing else, and no list is sold or shared.