Start with the AI Gateway
Find out what AI your organisation already uses, and what it costs. One governed endpoint in your perimeter that every coding agent and application points at instead of a vendor, your tools unchanged — and the first spend report, per developer, a week after it goes live.
- Monthly fee
- Starts at €500 / month
- Duration
- Deployed in one day
- Who it needs
- A platform lead, and the developers who already use AI
This is the one part of the platform you can adopt on its own. It needs no ontology, no process flows, no data integration and no programme: a base URL and a key, and the tools your people already use keep working unchanged. Behind that address, every call runs as the person who made it, against the models you allow, screened for what must not leave, metered against a ceiling, and written to an audit trail. Vendor keys stay on the platform; none is ever issued to a laptop.
Seats are becoming tokens because that is the right economics for agents — an agent that re-sends its whole context every turn is not a seat. It is also an open-ended line item unless you hold a control point of your own. The gateway is that control point, and it is yours: swap the model behind it by configuration, keep the identity, the policy and the audit trail whatever you do next.
How deployment goes
- Morning
Installed where production would run
Your cloud tenancy, on-premises, or a European sovereign cloud — a small cluster of its own, sized for your headcount, that the rest of the platform can later grow into, so nothing has to be re-proven. Federated to your own identity provider, so a gateway key is minted against a named person and revoked by removing them.
- Midday
The first developers routed
A sign-in and a key each, then a base URL in the environment. Claude Code, Codex, Gemini CLI, Cursor, Aider and anything else that speaks a standard completions API runs governed from that moment — no plugin, no wrapper, no forked client, prompt caching preserved.
- Afternoon
Your models and your contracts behind it
The vendor APIs you already pay for, models in your own cloud tenancy, or open-weight models served on your own hardware — reached through the same endpoint and chosen by name. A model that is named is the model that runs; nothing is ever silently substituted.
- End of day
Policy, once — and live
Which models each clearance may reach, what data loss prevention screens on the way out, and which integrations every developer's agent may use — enabled once for the organisation, picked up on the next turn, revoked with the same gesture. The monthly cap is set with you before we leave, and the gateway is live.
A week after go-live, the first spend report
Deployment takes a day; the report needs a week of real usage, and a week is enough. Spend, calls, tokens and carbon, attributed per developer and per team or unit, and broken down by model, capability and workflow, with an on-pace projection for the month against the cap you set. This is the document a champion forwards, and the number finance did not have.
What the first report looks like: spend, calls, tokens and carbon for the month, per developer and per model, against the ceiling you set — with an export for when finance asks.
What you get
An inventory you did not have
Who uses which model, for what, how often. Shadow AI turns out to be demand, and now it is demand you can see, meet and govern.
A number for the token contract
What a month costs on consumption pricing, per team or unit as well as in total, projected from real usage rather than a vendor's estimate — and a cap so the invoice cannot surprise you.
No vendor key on any laptop
Every call runs as a person, resolved through your identity provider. Remove the person and the next turn stops.
Governance that costs milliseconds
Allowlist, data loss prevention, moderation, metering and audit on every call, at single-digit millisecond gateway overhead — less than a cold TLS handshake to a vendor.
Developers who noticed nothing
Same tools, same commands, same muscle memory. The governed path is the easy path, which is the only kind of governed path people stay on.
A model you can change
Frontier API today, open-weight on your own hardware tomorrow, swapped by configuration. The gateway is the thing that stops a vendor locking you in.
What it costs
- A fixed monthly fee for the cluster that carries your people — not per developer, not per token
- Excludes hosting and model spend: the cluster runs on your infrastructure, the models on your own vendor contracts
- Credited in full against your first platform year if you expand
The gateway is priced on capacity, not headcount: a fixed monthly fee, on a yearly commitment, for the cluster size that carries the people who will use AI through it. The smallest cluster carries 150 users; the largest we price on this page carries 6,000. Because the fee is for the cluster, adding a developer is a sign-in rather than a line item, and usage never changes the invoice — the gateway attributes and caps your model spend, which stays on your own vendor contracts. The fee is for the gateway itself and excludes hosting: the cluster runs in your cloud tenancy, on-premises or on a European sovereign cloud, on infrastructure you pay for directly, and the deployment day is included. All prices exclude VAT.
- Cluster
- S · up to 150 users
- Per month
- €500
- Per year, on a yearly commitment
- €6,000
That works out at about €3 per user per month, but nothing is billed per user: the fee is for the cluster, and adding a person is a sign-in. Hosting is not included — the cluster runs on your infrastructure — and your model spend stays on your own vendor contracts, attributed and capped by the router.
What comes after, if anything
Nothing has to. Plenty of organisations will run the gateway on its own for a year and get exactly what they came for: one address, one invoice they can read, and vendor keys nobody has to rotate. If you later want governed agents, tools over governed MCP or grounded retrieval, the route, the identity, the policy and the audit trail are already in place — the expansion is a continuation, not a restart, and the gateway fee is credited against it.
Where this sits
- Deployed in one dayAI GatewayStarts at €500 / month
- 30 minutesOrganisational AI in 30 MinutesFree
- One day eachThe assessment days€1,500
- Six weeksThe Six-Week Pilot€25,000 – €90,000
Every other rung asks something of you first: a room full of process owners, a data source, a sponsor, six weeks. This one asks for a base URL. It is where to start if the question in front of you is what is our AI actually costing us and who is using it — and if the question is bigger than that, the briefing costs nothing and says so.
Questions people ask before they say yes
Do developers have to change how they work? No. A sign-in, a key, a base URL in the environment. Their existing tools run governed from the next command, with prompt caching preserved.
Which models can sit behind it? Frontier vendor APIs, models hosted in your own cloud tenancy, and open-weight models on your own hardware — through the same endpoint, chosen by name, swapped by configuration. If a request names a model, that model runs or the request fails with a typed error; a cheaper model is never substituted behind a developer's back.
Who does a call run as? The person who made it, resolved through your own identity provider. That is what makes per-developer cost, clearance-gated model access and instant revocation possible, and it is what a gateway that merely fronts your keys cannot offer.
How is this different from a model-only gateway? Such a gateway governs the model call, and the good ones do it well. This gateway shares one identity, one policy and one audit chain with the platform's governed tool calls and the sandbox those tools run in, so governance does not stop where the model call ends. The full comparison is on the concept page.
Can we run it air-gapped? Yes. The same cluster runs disconnected, with open-weight models behind it; that is a deployment choice, not a different product.
What do you need from us before the day? Someone who can grant an identity-provider application, the vendor contracts you want behind the gateway, and a platform lead who owns the cap. We send a one-page prep note when we confirm the day.
Prefer to write? Email hello [at] scrydon.com and we will get back to you.