Agents operate. People approve.
A self-service aggregator run by agents, yours and ours. A person approves anything touching money, keys or contracts.
What we are building
Our bet: within a year, anyone with traffic can open a casino in a couple of days without an engineer. What they will not have is the games. That is the gap we are building for.
A casino in days, for anyone with traffic
Platforms are getting to the point where a webmaster or an affiliate with an audience and no technical background can stand up a casino in a couple of days. We think that is where the market is within twelve months.
The games are the missing piece
A casino with no games is a landing page. Getting games today means an aggregator, a sales call, a contract, an integration measured in months and a team to run it. The people opening casinos next year will have none of that.
Games from the agent window, in one window
We are building for that operator: they ask their agent for games, and the agent does the rest with us. Catalog, wallet, test spins, readiness, go-live, in the window where the casino itself is being built. Today it takes one engineer and an afternoon. The target is no engineer at all.
Everything that can be automated, will be
Onboarding, integration, monitoring, support, announcements, billing: we are moving every process onto agents and keeping people for the approvals that touch money, keys and contracts. That is what lets a small team serve every casino that opens, and what keeps the price a fixed fee per spin.
How the platform is run
Legacy aggregation is sales-led: a call, a quote, an integration team, months. That model runs on headcount, and the headcount is paid for out of a share of your revenue. This platform is built the other way round.
Agents do the operating
Integration, monitoring, partner chats and announcements run on automation, around the clock. Sign-in checks run on rules. Your agent does the integration. Ours run the platform.
People approve
Money, keys, contract terms and go-live are decisions, and a person makes them. Not as a fallback when an agent is unsure: as rules written into code. They are listed below.
Skills instead of staff
Each agent is a written skill: what it may say, what it may touch, when it must hand over. A new capability is a new skill, not a new department. That is what keeps the team small.
Knowledge is governed before it is used
The assistant drafts only from a short set of facts a person has reviewed. A fact nobody reviewed is a fact the agent does not know, and a question about it goes to a person.
Where the human sits
Not a principle. Six rules, each enforced by code or written into the contract you sign.
Money moves by code, never by a model
You prepay a package and code meters every spin. Hosted MCP is read-only; possessing an operator key does not enable its action tools. Authorised HTTP session requests use the real wallet flow, including when welcome credits cover platform usage.
A person with the right role issues the key
An owner or admin of your organisation issues one API key in the cabinet or Mini App. In the Mini App the step sits behind your second factor if you have enrolled one. Store the key in a secret manager and configure your backend or client locally. You can revoke it yourself.
A launch is announced by people, then agents
The provider submits a launch announcement. Aggregator.gg approves it and shares it with operators. The games launch when the provider goes live.
A person lets the bot into each chat
Every partner chat is bound to one company by a human decision in our admin. Two bindings or none, and the bot stays quiet rather than guess whose chat it is.
A pasted secret never reaches the model
An API key or password pasted into a partner chat is cut out before the message is stored or shown to a model. What stays is a fingerprint: enough for a person to see which key it was, not enough to use it.
The model advises, code decides
Whether the bot answers, escalates or stays silent is a ladder of rules, not a mood. A model can talk us out of answering. It can never talk us into a promise.
What that buys you
Under ten people run the platform. Agents do the operating. People approve. There is no account manager you did not ask for, no quarterly renegotiation to staff, no integration team you have to wait for. So the price does not have to be a share of your revenue. You pay for spins. You do not pay for our overhead.
What runs today
Each of these is in production. Each names the point where a person decides.
Your coding agent does the integration
Connect Claude, Cursor or another MCP client for read-only catalog and documentation discovery. Your agent can help write the integration code. Configure callbacks in the cabinet and use the authorised HTTP API for session requests. Your team reviews and validates the integration before moving player traffic. What the catalog holds
The readiness check attacks your wallet first
Seventeen scenarios run against your integration: forged and missing signatures, a duplicate debit, the same transaction replayed with a different body, zero and negative amounts, a currency mismatch. The report is advisory evidence for your team. Key issuance is available before these checks; configuring API access is not a readiness pass.
A monitor watches live games on every operator
Every six hours it compares today's spins with the same weekday over the past four weeks, per game and per operator. A drop of a fifth or more is written up as an alert for our team and the provider's. The agent flags. It never touches your configuration.
Partner chats: an agent drafts, a person sends
Our assistant sits in partner Telegram chats. It drafts answers only from reviewed facts, and a draft reaches the chat only after a person on our side presses Send. Integration mechanics, price, limits, contract terms, release dates and incidents go to a person, every time.
Nothing reaches your partner chat unauthorised
An assistant may draft a message. It is delivered only to a chat a person has bound to your company in our admin. No binding, no delivery, and the rejection is recorded.
Every sign-in is scored by rules
A new address, a new browser, a new country or a run of failed attempts each add to a risk score on the login. A sign-in from a blocked jurisdiction is refused outright. The score is on the record, and a person can read why.
One engineer, one afternoon, no integration project
Crypto Casino Operator runs a Telegram-first crypto casino. After lunch, their engineer pointed Cursor at our MCP server. By evening, the casino was live.
That is the whole model, compressed into an afternoon. An agent did the wiring; a person decided.
What we do not automate
RTP deviation, RNG certification and bot detection belong to the game provider under the offer they sign with Aggregator.gg. Your player wallet, your GGR and your settlement are your ledgers, not ours. We route spins and we bill for them.
An AI-first aggregator that claimed all of the above would be claiming to run your casino. We are not. We are the layer between you and the studios, run by agents, with people where people belong.