The blueprint, in one page
What we are building, who it is for, what an agency can do with it, and what it costs. No case studies, because there are none yet. If you would rather argue with the premise than read a pitch, the premise is in the first section.
What we found looking at 1,758 agency sites
Before building anything we scanned reachable agency websites in 12 languages and checked, with quoted evidence from the page, whether they sell GEO or AI-search services. The gap between capability and offer is the whole thesis.
Method note: one cohort was sourced by searching for agencies that already market GEO, so it is excluded from every denominator above. The Italian and international cohorts differ sharply, at 8% and 29% respectively, so treat the blended figure as a range rather than a single market rate.
The readiness half of that scan is something you can reproduce on one domain at a time: check whether a client site is readable by AI crawlers and you get the same nine pass or fail lines we counted, with the fix for each one.
The problem
Not "agencies need more content." Four specific things that stop an AI-search offer becoming a billable line.
Clients are asking, and there is no repeatable answer
The question arrives before the offer does. Most agencies can explain what AI search is, but cannot hand a client a documented method, a scope of work, and a monthly report. Without those three things it stays a conversation rather than a line on an invoice.
The measurement tools disagree with each other
Point three AI visibility tools at the same brand with the same prompts and the citation rates come back different. Because each vendor keeps its methodology private, there is no way to tell which is closer to the truth, which makes the number very hard to defend in a client meeting.
Volume does not scale across a roster without headcount
Producing content for one brand is a workflow problem. Producing it for fifteen, each with its own voice, CMS, and approval chain, is an operations problem. The usual fix is hiring, which puts the margin back where it started.
Unreviewed AI on a client brand is a real liability
Google's 2025 scaled-content enforcement made high-volume unreviewed AI a penalty risk. On a client site that risk is not yours to take. Most tools in this category still default to publishing automatically.
The approach
Four decisions, each one aimed at a specific problem above.
It is the execution layer, not another dashboard
Research a keyword, build the brief, write the draft, score it, and publish it to the client CMS. The output is published pages, not a report telling you what to go and do.
Open source, the whole product
Self-host it on your own infrastructure. No per-seat pricing, no procurement cycle, and no vendor who can change the terms once your clients depend on it. Multi-client workspaces and white-label reporting are in the free build rather than behind a premium tier.
Nothing publishes without a named person behind it
The approval gate sits inside the publish step, so scheduled jobs and bulk actions cannot route around it, and publishing defaults to draft. You can sell editorial judgement rather than automation, which is the safer thing to put on a client brand.
The measurement is auditable
The prompt set, the sampling method, and the scoring checks are readable source rather than a black box. When a client asks how the number was produced, you can show them, which is a materially different conversation from citing a vendor dashboard.
The outcome, stated as what you can sell
- A GEO or AI-visibility retainer you can scope, price, and defend, built on a method you can show the client.
- A monthly client-facing report under your own brand, on work the system actually did rather than third-party data re-displayed.
- Content throughput across the whole roster without adding writers, with per-client voice and 10 CMS destinations.
- A software cost that is your infrastructure bill rather than a per-seat line that grows against you as you win clients.
We are deliberately not putting a revenue multiple here. We have no customer results to base one on, and you can price your own retainer better than we can guess at it.
How the agency layer works
The parts that exist because a roster is not one site repeated. Workspaces, branding, roles, and the approval trail, described as they are actually built rather than as a feature grid.
One workspace per client, and workspaces are not the meter
Each client gets an isolated workspace carrying its own site, keyword set, voice profile, and publishing destination. Nothing is shared between them unless you share it. The Agency tier does not cap how many you run, so taking on a client costs nothing until that client’s articles are actually produced. Managed stops at three workspaces, which is the honest line between one operator and a roster.
White-label is ungated, including in the free build
The agency name, accent colour and the address monthly reports are sent from are account-level settings, applied to everything a client sees. This is not an upsell: under the licence there are no feature gates, so the self-hosted build carries the same white-label and multi-tenant capability as the paid rungs. Anyone selling you client-facing branding as a premium tier is charging for something we do not gate.
Three roles, so a writer cannot rebrand the account
Owner, admin, and editor. Editors draft, review, and approve content but cannot touch billing, branding, or the client roster, and changing white-label settings is restricted to owner and admin. That is the split that matters once a junior writer is working inside a live client brand rather than a sandbox.
Approval sits inside publish
The gate is part of the publish step rather than a status field beside it, so scheduled jobs and bulk actions cannot route around it, and publishing defaults to draft. Add the client as a reviewer and the trail records who approved which version and when, which is the artifact you want when a client asks why a page went live.
Per-client voice, eleven publishing destinations
Every workspace trains its own voice profile from that client’s existing writing, so output does not converge on one house style across the roster. Publishing targets WordPress, Shopify, WooCommerce, Magento, Webflow, Ghost, Wix, HubSpot, Notion, Framer, or a git repository, which covers most of what a mixed client base actually runs.
Put in roster terms: the Agency tier's 400 included articles is roughly 26 a month across a fifteen-client book, or 40 across ten, before a single overage euro. The constraint you will hit first is editorial review capacity, not the plan, which is the constraint we would rather you hit.
The price
Metering is on articles published and on who pays the model bill. It is not on seats and not on how many client brands you run, because charging an agency for winning clients is a bad way to be an agency's tool.
Self-host
The whole engine, open source. Your infrastructure, your API keys, no feature gates.
Managed
No API keys to manage, because model and data costs are included. For solo operators and agencies running one or two brands.
Agency
For agencies running content across a full client roster. Everything metered on output, not seats or workspaces.
Full breakdown on the pricing page, and the licence detail on the open-source page.
Where this actually is
- Pre-launch. No paying customers, and no case studies, which is why there are none on this page.
- The repository is not public yet. The licence and scope are decided, the code is not out.
- Early engagements are being lined up now.
- The repository is public and carries the whole product, dashboard included, with an MCP server. A packaged CLI is not built yet.
- The agency scan above is real, first-party, and we will share the method with anyone who asks.
If any of the four problems above is one you actually have, the useful next step is twenty minutes on how your agency scopes and reports this work today. Not a demo. There is nothing to demo yet, and your answer changes what gets built.
Start that conversationThe obvious questions
Is this available today?
Yes. The repository is public and you can self-host it now. The hosted tiers are also live.
What are you actually asking for?
A conversation, and for a small number of agencies, a design-partner relationship. That means you tell us how you currently scope, deliver, and report AI-search work, we build against that, and you get early access and direct influence over what ships. It is not a purchase and there is nothing to sign.
Where do the numbers on this page come from?
From our own scan of 1,758 reachable agency websites across 12 languages, checking whether GEO or AI-search services appear on a services page with quoted evidence from the page itself. One cohort was sourced by searching specifically for agencies that market GEO, so it is excluded from every denominator here. It is first-party data, and we will show you the method if you want to argue with it.
What does the licence mean for my agency?
Self-hosting it and running it as-is for your own clients carries no publication obligation. Using the software commercially and on behalf of clients is exactly what the licence permits. The obligation only triggers if you modify the source and then offer that modified version to others over a network. Running an agency on it is free and unencumbered.
Do you charge per client workspace or per seat?
Neither. Metering is on articles published and on who pays the model and data bill. The Agency tier at €199 a month includes 400 articles and an unlimited number of workspaces, with role-based permissions for your team included rather than priced per user. Winning a client should not increase your software bill before that client has produced anything.
What happens if I publish more than 400 articles in a month?
Each additional article is €0.45. There is no forced jump to a higher tier and no overage penalty, because the marginal cost is a real model and data cost rather than a pricing lever. On Managed the same overage is €0.60, so the Agency rung lowers your marginal cost as well as raising the included volume.
Is white-label reporting locked to the Agency tier?
No, and this is worth being precise about because most tools in this category do lock it. White-label branding and multi-client workspaces are ungated, including in the free self-hosted build. The paid rungs sell hosting, included model and data costs, volume, roles, and support, not access to client-facing branding.
Who approves content before it goes live?
A named member of your team. Approval is the editorial gate: nothing is published until someone with access to that workspace approves it, publishing defaults to draft either way, and every article records who approved it and when. There is no separate client-facing sign-off step today; if a client needs to review, give them an editor seat on their own workspace.
Why would you give away the parts other tools charge for?
Because the code is not the moat. Competitors have shipped the same features and the category is well served. What we do not have is distribution or a reason for an agency to trust a tool with no track record. Shipping everything open is how we buy both, and the licence is what stops a competitor from taking it and closing it.
Be the site the assistant names.
Add a domain and the first draft is written while you watch. Every publish is your decision, a click on the draft or a rule you set and can hold, and you can read the source or self-host it free.
Add a domain, it sets up your workspace