Waitlist
Join the build-in-public waitlist.
We are building Quarrybank in the open. Leave your email for early access, and the write-up when the first real dogfood numbers land.
Quarrybank. The agent does the work. You approve.
Same data the big tools use, minus the dashboards, plus the engineering that makes it trustworthy. Quarrybank pipes best-of-breed market data into the AI chat I already work in, and the agent finds the angle, writes it in a human voice, gates it, and queues it to publish. I approve.
"On a budget, the best LED grow light under £40 balances PAR output against running cost..."
One operator, producing like a content team, because the agent does the work. I am running my own publisher sites through Quarrybank this month, in the open. Every article has to clear a hard, independent quality gate before it can be queued. The real figures, how many it produced and how many are already cited in AI Overviews, land right here as the month runs. Numbers, not adjectives, and never before they are true.
Every SEO tool shows you charts, then leaves the work to you. Semrush and Ahrefs have a real moat, their data, but the dashboard is not it. Quarrybank keeps the best-of-breed data and throws away the dashboard, piping the raw power into the agent that already does your work.
Best-of-breed sources, live SERPs, search volume, AI-Overview citations, backlinks and your CMS, piped straight into your AI chat.
The agent scores what is actually winnable, and where the advertiser money is, before you write a word.
It writes in a real human voice and runs every piece through the same independent quality gate. No marking its own homework.
You approve, it ships to your site. Publishing always stays human-confirmed.
Under the hood: many data sources, each with its own context store, into a unified infra layer, into one MCP server, into your AI chat, shipping to your site, every piece quality-gated. See the architecture →
On a modern global edge. Fast everywhere, nothing to babysit. No server patching, no plugins to maintain.
The full pipeline: idea, research, draft, quality gate, publish.
Your topical map, source library, competitor and entity maps, built and kept current.
Research, winnability, citation-gap scouting, value-pocket finding, human-voice writing, quality gating. Each loaded only when needed, so it stays fast and cheap.
Six to ten connected: SERPs, search volume, AI-Overview citations, backlinks, Search Console rank truth, affiliate and social.
Request a page from a prompt and it appears as a gated draft on your site.
Branded /products/ links that redirect to your affiliate URLs and track clicks, built in.
Your keys in a real secret store, isolated and rotated, never in plaintext.
They show you data. Quarrybank's agent does the work with it. The analysis arrives where you already work and turns into published output, with no context-switching into a dashboard.
No server, no plugin soup, no patching. And the detail that matters: in WordPress an agent with write access is code running with broad database access, a real liability. In Quarrybank every AI action is a constrained, sandboxed, gated request, and you approve every draft.
Every output passes an independent quality gate, the same judge every time, never grading its own work. Real tests, real deploy pipelines. That is why you can let an agent run a pipeline without it quietly shipping rubbish.
Find, write, gate, queue. The whole motion runs from the AI chat, with a human approving before anything ships. A short screen recording of that flow is worth more than any paragraph, so it goes right here.
Following along is the point. Read the build log →
Most people, given the same goal, never arrive here, because it sits where five disciplines meet, and few hold all five at once.
Git, CI/CD, real tests, deploy pipelines, one-click provisioning.
Independent, one-judge evaluation instead of an agent grading itself. Skills that carry their own context, loaded on demand. Harnesses for long-running agents.
Genuine human voice, answer-first structure, honest affiliate practice.
Winnability scoring, AI-Overview citation gaps, equivalent search value.
MCP-native. You operate from chat, not a dashboard.
Each layer is copyable. The integrated, harnessed combination is not. That is the moat.
Join the waitlist free today. A blueprint pack for builders who want the whole rig, a managed turnkey for teams who want it stood up for them, multi-tenant for agencies, and a source licence if you want to own it outright. We label what is live now and what opens as the one-click provisioner ships, no turnkey promises before the button exists.
See the full offer ladder → · The blueprint pack → · Enterprise →
No. It is the data integration, the engineering harness (independent quality gates, tests, provisioning) and the agent workflow. The prompt box is the smallest part.
For the blueprint pack, yes, it is for builders. For the managed tier, no, we set it up for you.
Supported as a publish target. Most people are better off on the AI-native CMS though: no server, no plugins, no patching.
Keys live in a real secret store, per-tenant isolated and rotated, never in plaintext.
You can own the source outright with a licence. It is built on open standards (MCP) and best-of-breed data you could take elsewhere.
No. Every piece passes an independent quality gate, and publishing is always human-approved.
We are building Quarrybank in the open. Leave your email for early access, and the write-up when the first real dogfood numbers land.