Your Brand, Callable

Here is the transaction, all of it, in four sentences. An AI agent you already use asks Jinn for your brand. What comes back is the verified record: your voice register, your palette, your positioning, the claims you're allowed to make. The agent writes from those facts instead of guessing. The call lands in your audit log with your own text scrubbed out of it, and the agent gets on with the work.

One key connects the assistant in your editor, the agent running your inbox, or any tool that speaks MCP, the shared standard agents use to plug into a service, to that record. Paste one block of configuration and the agent is grounded. Your brand, callable by any AI agent, is the product's own headline, and this piece is about why a brand needs that, what an agent can and can't do once it has it, and what a small team stops doing by hand.

Why a paste isn't enough

Most people's first attempt at this is to load the brand book into an AI tool and type "understand my brand." I've done it. The result does a passing job. It holds the tone for a paragraph, it gets the product name right, and for a first draft of something generic it's fine.

Then you push it. Ask for the fourth variation. Ask for the email to the customer who already bought, in the voice you'd actually use with someone who trusted you once. Ask for the ad aimed at the audience you haven't reached yet, and watch it fall apart, because what it had been holding all along was a summary, and a summary is what you get back.

A brand is almost entirely details. Jinn maps hundreds of them, 346 signals per brand, from the customer's own voice to every component of the brand and how those components intersect, and feeds them into an agent's work the way you'd feed them into a planning session with your team around a table: which claim is allowed, which word is banned, what the accent color never sits on, what the positioning is against, and which competitor the agent must never name as a comparison. The depth is the difference, and it compounds: the record learns from what works and from your feedback, so the agent that reads it after a year of your feedback reads a better brand than the one that read it on day one. The honest side of that is where it starts. A young record holds less than an old one; the first weeks are spent confirming facts, and the agent is only as deep as what you've confirmed.

The record also knows what it doesn't know. An agent can ask the brand a plain question and get back the facts that bear on it, and where your record can't answer, the agent says so, with the gap named. An agent that knows where your brand is thin stops inventing the thin parts.

What an agent can do with it

Twenty-two tools, and every one of them reads the same facts a person logged into the product would see. Nothing is rebuilt for the agent, and nothing is a mockup standing in for the real thing.

Almost all of them read. An agent can read your Brand DNA and your products one by one, your design system as a file it can build from and as the raw tokens behind it, your brand kit, your voice profile, the status of your strategy work and what it's delivered, what the AI engines currently get wrong about you, the inbox where new facts about the brand wait for your approval, and the set of things your brand has decided are true. A coding agent can build an on-brand page without a human pasting a style guide, and a writing agent can transform a draft into your voice. A research agent can check its own claim against your record before it makes it.

Two tools write, and neither can spend. One queues a checkout link or an intake form and hands control back to you. The other files a document into your learning inbox, where it waits until you approve it. An agent can prepare a purchase; a person completes it. There's no setting behind that. The permission doesn't exist, so you can't grant it by accident. The two writing tools were built unable to complete a payment, which is the only version of agent spending I'd put my own brand behind.

The key itself is worth understanding, because it's where trust lives. It's tied to specific brands of yours and nothing else. It checks your subscription every time it's used, so if you upgrade, renew, or lapse, the agent's access changes on the very next call, with nothing to re-issue. If the system can't confirm your standing at that moment, it refuses. Every call is capped in how fast it can come and logged, with the free text, your prompts and your copy, redacted before the log is stored, so the trail can't be reverse-engineered back into your content. You see the key exactly once when it's created, and you can rotate it or revoke it from your account whenever you like.

How it fits the week you already have

Your tools stay where they are; the briefing they start each session from is the part that changes.

Before the key exists, every agent your team runs starts each session from a blank prompt, and the brand gets re-briefed into it by whoever is typing, from whatever they remember. The landing-page agent gets a screenshot of the old page, the email agent gets last month's email, and the research agent gets nothing and confidently fills the gap. Three tools, three slightly different brands, and a person spending the first stretch of every session re-teaching what the last session already knew.

With the key in place, the briefing is the record. The same three agents from a minute ago now read one record: the landing-page agent pulls the design system, the email agent pulls the voice profile with its banned words, and the research agent checks a claim against the record before it writes it. When something about the brand changes, you tell the record once, and every agent that reads from it picks the change up the next time it's called. We run Jinn this way ourselves; the company is its own first customer.

I'd rather name the limits than have you find them. The record only answers what it holds, which is the young-record point from earlier, and it holds more every week you confirm something. The two writing tools hand work back to you instead of finishing it, which means a person is still in the loop on every purchase and every new fact, on purpose. If what you wanted was an agent that spends and publishes unattended, this is the wrong product, and I'd argue it's the wrong instinct.

What it replaces

Start with the re-briefing, the paste at the top of every session and the re-teaching that follows it, which is small each time and enormous across a team and a year. Then the drift that re-briefing produces: three tools running three versions of the brand, none of them checked against anything.

And a class of tool that looks similar from a distance. Connector gateways give an agent a thousand generic app actions, routed through to other companies' systems. Jinn Agents gives it your brand: tools that return your verified facts, checked against the same record every Jinn product runs on, with the spending line drawn by construction and the access line checked live on every call. A connector answers what the agent can do. The bad draft came from the other question, what the agent knows about you, and the tools here exist to answer it.

Try it without an account

You don't have to take any of this on faith. A free demo key needs no login and works read-only against a small set of showcase brands, for a limited window. Point the agent you already use at one of them and ask it to write something. Then ask it to write the same thing with the key removed, and read the two side by side. The difference between them is the product.

Jinn Agents is in early access. Asking the brand a question is available from the Connected tier up.