Skip to the content

Every agent Kwern runs

One agent per channel, 14 of them. Each reads your real data on a fixed cadence and drafts the work that belongs on its channel: a reply, a post, a listing change, an article, a bid. You approve what goes out.

Get started

Bring your own model key. Nothing publishes without your approval.

Reads and writes through the services you already use

App Store Connect
Apple Ads
Google Play
RevenueCat
Stripe
PostHog
Google Search Console
Google Analytics 4
Google Ads
X
Reddit
LinkedIn
GitHub

What is an AI marketing agent?

An AI marketing agent is a program that does one channel's work on a schedule: it reads what is happening on that channel, decides what is worth acting on, and writes the thing a person would have written. It is not a chat window you have to open, and it is not a scheduler that publishes whatever you queued last Sunday. The reading is the part that takes the time, and the reading is the part it does for you.

In Kwern an agent runs as a Loop. A Loop wakes on its own cadence, reads the product's real data through the connectors you have given it, runs the marketing skills its template allows, and leaves what it produced in the approval feed. Some wake often because their channel moves fast. Some wait longer because a site does not change overnight. The ones that spend money leave the ad platform time to learn from each change, and pace themselves against a monthly budget you set.

The cadence is the point. A channel agent that you have to open is a chat window with a job title, and it goes the way every other tool you tried went: three good weeks, then a tab you stop opening. An agent that wakes on its own and leaves the work where you will trip over it keeps producing on the weeks you are shipping, which are the weeks marketing usually stops. What it costs you on those weeks is the two minutes it takes to read a draft and say yes or no.

Which agents you get

You do not pick from a menu. Kwern reads the URL you give it and works out what the product is: an App Store listing, a website, or the same product on both. The agents that need a surface you do not have never appear, so nobody has to decide whether an App Store agent is worth turning on for a product with no app.

After that the only decisions left are the ones that cost something. The two ad agents and the influencer agent are off until you turn them on, because one spends your money and the other writes to people. The Writer is off too: an article nobody publishes is a day of the model's time for nothing. Everything else runs from the first morning.

What one run does

  1. 1

    It reads first

    Through the connectors you gave it, and the public threads of Reddit and Hacker News. The X agents search the day's posts, the Reddit agent recent threads, the Hacker News agent the week's, the SEO agent crawls your site, the ASO agent reads the listing and the week's reviews, the ad agents read the campaigns and what they spent. An agent missing the connector it needs does not run and does not pretend to: it says what is missing.

  2. 2

    It runs the skills its template allows

    Only the ones its template allows, entry skill first. The Reddit agent cannot decide to rewrite your App Store subtitle and the ads agent cannot decide to write a blog post. That list is on this site, one page per skill, and you can read the file before the agent runs it.

  3. 3

    It writes down what it was looking at

    The draft, and the reason for it: the thread it answers, the keyword it targets, the search terms behind a bid, the review it replies to. A draft with no reason attached is a draft you have to take on faith, and taking marketing on faith is how the last tool went.

  4. 4

    It leaves the work where you will see it

    Most agents leave a draft in the approval feed. The SEO, GEO and Competitors agents leave something else: a recommendation, a pull request on the repository that serves your site, or a digest of what your competitors changed. Neither kind goes anywhere until you have read it.

What an agent never does

The four limits that hold on every channel, whichever agents you have switched on.

It never publishes or spends on its own

Approve is the only way anything reaches a channel or an ad account. There is no autopilot setting, no scheduled send, and no mode where an agent spends first and tells you afterwards. Where a service does not let Kwern write at all, approving is not even a button: you get the text and the link, and you post it yourself.

It never opens something new on its own

The ads agents propose bids, budgets, negatives, keywords, pauses and, on Google Ads, the campaign target. They cannot create a campaign or an ad group: new campaigns start in a chat you open, one proposal per campaign, created only when you publish it. Their instructions keep each move within a fifth of the current value unless they say why, and every run reads the monthly budget you set. A mistake is a small mistake that the next run can undo.

It never pretends to be somebody else

An agent posts as you, from your own connected account, or it does not post. It never runs a second account, never writes to somebody who has not been shown to you first, and on Reddit and Hacker News it never posts at all, because an automated account on either is an account with no standing.

It never makes a number up

Every draft is written against the foundation document Kwern wrote from your own pages, and against the data your connectors returned. The Influencers agent, which writes to a person, goes further: what only you know is left in brackets for you to fill in rather than invented because it reads well.

What every agent has in common

Nothing leaves without your yes

An agent proposes. It never publishes a post, changes a listing or raises a bid on its own, and there is no autopilot switch hidden in the settings. Drafts land in one feed, and you approve, edit or throw each one away.

The model runs on your key

The model runs on your own provider key, at the provider's price, on your account. Kwern records what each run cost and never resells a token. Across the 11 models it vets, input runs from $0.10 to $10 per million tokens, and which end you sit at is your call.

The skills are open

Every agent runs the open marketing skills in the catalogue, 75 of them, most taken as they are from Corey Haines' marketingskills and makerskills and credited to the commit they were vendored at. You can read every one before you turn an agent on.

You decide which ones run

9 of the 14 are on the day you sign up. Four are not: the two ad agents, because they spend your money; the influencer agent, because it writes to people; and the Writer, because an article nobody publishes is a day of the model's time for nothing. Any of them can go back off in one click.

Three plans, every agent in each

Your model key, your accounts, your approval on every post. The plan decides how many products you run.

MonthlyYearly-17%

Lite

/ month

Billed monthly

Includes:

  • 1 Product
  • Usage included
  • Every agent, every skill and every service you can connect

Pro

/ month

Billed monthly

Includes:

  • 3 Products
  • 3× Lite's usage
  • Every agent, every skill and every service you can connect

Max

/ month

Billed monthly

Includes:

  • 5 Products
  • 5× Lite's usage
  • Every agent, every skill and every service you can connect
What the subscription covers

Someone is asking for what you built today.

On Reddit, on X, in ChatGPT's answers. Kwern finds them every morning and drafts the reply in your voice; you decide what goes out.