Managed AI operations

We keep your AI agents working — and we are on the hook when they don't

Most AI projects do not fail on launch day. They fail in month four: the vendor changes the model version, output quality slips without anyone noticing, and the person who built it has moved on. By then your team has quietly gone back to doing the work by hand — while you keep paying for both.

We run your agents in production under a service level agreement, with contractual responsibility for them working, not just for delivering them.

What you get every month

What the SLA actually commits us to

Availability, incident response times by severity, a minimum output-quality threshold below which we investigate at our cost, reporting frequency, and the delivery date of your compliance evidence. All measurable. If we miss them, it shows up in your own monthly report — you do not have to catch us at it.

Your people stay in control

Each agent has a confidence threshold. Above it, it acts; below it, a named person on your side gets the case with the reasoning attached. We calibrate that threshold during the pilot and tighten it as history builds. The point is not to remove your staff — it is to have them working only on the exceptions where their judgement is worth paying for.

This is also exactly what the EU AI Act expects for human oversight, so the same mechanism serves as your compliance evidence.

What is yours, and what stays ours

We build and own the agent framework, the prompt engineering and the reusable components behind it. That is our product — it is the reason we can do this faster and cheaper than you could build it in-house, and we do not hand it to the market.

What is unambiguously yours, always:

• Your data. Exportable in a standard format at any time. We do not resell it and we do not train shared models on it.
• Your audit trail. Every decision log, retained and exportable. This is your compliance evidence under the AI Act and NIS2, not our intellectual property — we will never hold it hostage.
• Your system documentation. What the agent does, what data it touches, how oversight works: everything your AI register needs.
• Your business rules, written down properly — usually for the first time.

Source code escrow is available for enterprise contracts: code, prompts and configuration are deposited with an independent third party and released to you automatically if we become insolvent, materially breach the contract, or discontinue the service.

That is the arrangement your procurement team is actually looking for. NIS2 obliges you to assess supplier dependency risk — escrow answers it properly, without either side pretending that handing over a vendor's product for free is a normal commercial term.

What it costs

A fixed monthly fee. The same number every month, no matter how much the agent processes.

We do not bill per token, per document or per model call. Model consumption is our problem, not yours — it is inside the fee, and we optimise it at our own expense. If your volume spikes in a busy month, your invoice does not.

That means you can put the number in the budget at the start of the year and stop thinking about it. No surprise invoices, no explaining to your CFO why last month cost three times as much. The fee is set after the pilot, once we know exactly what the process involves. Contracts usually start at 12 months.

Start with one agent, not five

We will not take on five processes at once, even if you ask. One agent, one process, running in parallel with the manual work until the numbers hold. Then expand.

Book a free 30-minute call — tell us what you would want running, and we will tell you what it takes to keep it running.