Your First Engine: Giving It a Purpose Your AI Tools Can Use

Most business software greets you with a settings form: a company name, a timezone, a logo upload, a wall of toggles you will never revisit. Context Engine asks one question first: What is this engine for? You answer in whatever words you have, and that answer ends up shaping more than anything else you write in the product.

A fresh engine is a blank memory. It has no customers, no decisions, no history yet. Soon, though, several AI tools will be saving to it, and some of them will reach more than one engine. Every time one of them has something worth keeping, it has to decide where it goes. The purpose is how it decides.

What a purpose is

An engine has no “kind”. There is no setting that says personal, team or client. Instead it has a purpose: a few sentences in your own words saying what it is for, what belongs in it, and what does not. A label like “team” or “client” would only hint at that; the purpose spells it out.

A purpose has three parts, and people usually skip the last one:

Our shared memory for Harbor Consulting's client work:
accounts, engagements, the people we work with, what we
decided and what we promised.
What belongs here: client meetings, decisions and their
reasons, commitments in both directions, contract and
pricing facts.
What does not belong here: internal hiring, personal notes,
other firms' work.

The company is invented; the shape is the one the product uses. The last line is what keeps an engine clean. Without it, an AI tool with a stray thought about your hiring plan and a connection to the client engine has no reason to put it anywhere else.

Where the purpose is read

The purpose is read in at least three places, and each one changes what your AI tools do:

• The brief. When ChatGPT, Claude, Claude Code or Cursor starts a session on a connection, the brief it receives opens with every engine the connection reaches and what each one is for, then what changed in each since that connection was last there.
• Choosing where to save. Every save names its engine. An AI tool that reaches a personal engine and a client engine reads both purposes and picks the one that fits. If none fits, it is told to say so rather than file it somewhere convenient.
• The consent screen. When you sign an AI tool in, each engine it could reach is listed with the first line of its purpose, so you can see what you are ticking.

With one engine, a vague purpose costs you little. With three (yours, the team’s, one per client), the purpose decides whether the memory stays sorted or slowly turns into a junk drawer.

How the first question goes

When you create an engine, the form asks What is this engine for? Type a sentence or two, as rough as you like. As you type, a suggested wording appears under the box: your words, put into that three-part shape, with nothing added. The engine’s name is suggested from it too, and you can change it.

Create engine uses the suggested wording. Use my own words instead uses exactly what you typed. If what you typed is too short to say what the engine is for (“acme stuff”), the screen asks one question instead, with three answers to tap, and Skip and use it as I wrote it does what it says. Nothing is saved until you choose.

The suggested wording is only a proposal. A statement about what your engine is for is a belief, and the product never saves a belief on your behalf without your yes.

Changing it later is in Settings › General, under Name and purpose, for an owner or admin. Saving takes effect at once, in the engine switcher, on the consent screen and in what your AI tools are told. If your words would read more clearly in the three parts, you get a suggested wording to accept with Use this version, or Keep mine. Renaming the engine never changes the identifier your AI tools use for it, so nothing needs reconnecting.

The website, if it is an organisation

Right after the purpose, the form asks for a Website. If you signed in with a work address, it is filled in from your domain. For a personal or project engine, tick No website and nothing is read.

Give it a site and Context Engine reads a handful of its pages while the engine is set up: the home page and the pages that say who you are, like About and pricing, not the blog, news or documentation. It keeps none of those pages. What it learned waits at the top of Home for the engine’s owner to check:

• About your organisation: the values your site names and how its writing sounds, each written as something your AI tools can follow.
• From your site: a few notes about what the site shows.
• Brand: the colours and logo it offered.

If the site gives itself a different name from the engine’s, that name is offered too. You go through them with Review, one at a time, and Keep, Edit, Skip or Later each; or Keep all. Nothing is saved until you keep it, and you do not have to answer before using anything else. All of it is a summary of your marketing, and if it were saved as fact, your AI tools would repeat it in your voice without anyone having checked it.

What you keep becomes ordinary memories: values and voice as identity memories that every AI tool can find, the notes as notes. For how to make voice do real work across every tool your team uses, see brand voice as infrastructure.

Where it all lives

The purpose, the values and the voice are all identity memories, stored the way everything in the engine is stored: typed, readable, linked. Three consequences worth knowing on day one:

• They have history. Every change records who made it, with which tool, and what it said before. A purpose that drifted can be traced.
• They can be edited in plain words. Open the memory and edit it, or ask a connected AI tool to change it. Editing the purpose memory changes the purpose everywhere.
• They go with you. An export of the engine is Markdown and JSON you can open without us, identity memories included.

Tuning it

Choice	Starting point	Alternatives
How many engines	one for what you are doing now	one for yourself, one per client or team
Purpose length	three or four sentences	shorter, as long as “Not” survives
Website	your organisation’s site	No website, for personal or project engines
Identity beyond purpose	values and voice from the site	add audience, positioning, principles later
Who can change the purpose	owners and admins	the same; members see it

Failure modes, and the fixes

The purpose is vague, so saving blurs. “General business stuff” cannot tell an AI tool which engine a note belongs in. Say what it holds and for whom, and say what it does not hold. A precise purpose now keeps several engines easy to tell apart later.

Memories land in the wrong engine anyway. Check the two purposes side by side; usually they overlap. Sharpen the “Not” lines, then move the misfiled memories. A moved memory takes its history with it.

You kept everything the website suggested. It happens with Keep all. Nothing is lost: open the memories under Memories and edit or archive what is not really you.

What changes

The immediate change is small. Your first AI tool’s first session starts with a brief that says what this engine is for, and its first save lands where it should.

The larger change is that everything you connect and save next, the meetings, decisions and files, goes into an engine that knows what it is for. The people you invite see the same purpose, so your team and your AI tools sort things by the same rule.

Build this yourself

1. Create an engine and answer What is this engine for? in your own words. Read the suggested wording and pick it, or your own.
2. Make sure it has a “Not” line. Edit it in Settings › General if it does not.
3. Give it your website, or tick No website for a personal or project engine.
4. Review what the site suggested on Home. Keep, edit or skip each one.
5. Connect your first AI tool, the last step of creating an engine, and ask it what it now knows. Then connect the rest.

See plans, or start free.