Your content operation, on autopilot.

Not yet, and we are not going to pretend otherwise. Sphinx is in build and opens to a first cohort, so what you get by asking now is a date, the cohort terms in writing, and a walkthrough of the engines it is built on. Two things make that a real offer rather than a waitlist. The engines underneath already run content operations for clients today, so this is a productization rather than a blank page. And tasken.ai becomes project number one inside it, onboarded through the same screens you would use, which is the gate we set ourselves: if it cannot run our own site as an ordinary project, it is not finished and we will not open it.

WordPress in version one, plus a universal export for everything else: the finished article with its title, meta description, schema and internal links attached, ready to paste or import. We would rather ship one connector that works properly than four that half-work. If your CMS is the blocker, say so and we will tell you straight whether it is next on the list or not on it at all.

One model key (an OpenRouter account covers everything, or bring a provider key directly), read-only Search Console access to your own property, and CMS credentials if you want it publishing for you. Search data and site crawls are on us, inside the monthly, so there is no second vendor bill to manage. The model spend is billed to you at cost by your own provider, never through us, and because you choose the models you set the ceiling yourself.

It writes from a brand brief derived from your own site: what you sell, how you talk, the proof you may use, the words you never use, and the call to action. You read it and correct it before anything is written, and we treat that screen as a gate, because a wrong brief quietly poisons every article after it. Every project carries its own brief, so an agency running fifteen sites has fifteen voices and none of them leak into each other.

You choose, per project, and you can change your mind either way. A new project runs on its own by default, because a machine that waits for a click on every article is a drafting tool. Three pauses survive whichever way you set it: you confirm the brand brief once at the start, pillar articles stop to ask you what only you know, and anything that fails the quality gate three times goes to a person. Full automation means publish what already passed the gate. It never means skip the gate, and the publish step structurally refuses an article that has not passed one.

It gets audited, and this is usually the fastest visible win. The moment Search Console is connected the engine reads your real coverage and positions and starts issuing verdicts on pages that already exist: two pages splitting one intent get merged, a page losing ground gets refreshed, a page sitting in positions 4 to 20 gets a plan to climb, and a page you rank with but that AI answers quote somebody else for gets restructured. Each verdict names the pages and the numbers behind it, so your editor can disagree with it before any work starts.

Yes. One flat $300 a month, however many projects, however many articles. Two honest caveats. The model spend is yours, so volume has a cost even though we do not charge for it, and you control that by choosing the models. And publishing two hundred mediocre pages is a worse outcome than publishing twenty good ones, which is why the gate exists and why we would rather you use the ceiling than fill it.

Neither, it is an AI System. We host the orchestration so there is nothing to install, and your team operates it (we do not run your content programme for you). The closest mental model is an AI employee, hosted for you, that your editor supervises.

Yes. Every prompt, every scoring weight, and the acceptance threshold itself are editable and versioned. Your senior editor can encode your standard into the system rather than filing a feature request and waiting. That is why we call it an AI System and not a SaaS: the logic is yours to change, not a black box you send tickets to.

A generator answers the question “write me an article about X”. The hard question is which X, and whether X should be a new page at all when you already have three that half-cover it. Sphinx spends most of its machinery on that decision: what to write, what to fix, and what to refuse, each with the evidence attached. The writing is the easy part now, and everyone has it. The judgement is the product.

Yes, that is Horus, and it is a separate system at the same flat $300 a month. Content and links are the two halves of the same job and most teams end up running both, but they are sold separately and you are not obliged to take one to get the other.

No contract, cancel any month, and access ends at the end of that month. Everything you built stays yours and exports in full: the brand brief, the backlog with its verdicts, every draft and every published article. Your model key was always in your own account, and anything already published is already on your site.