AI on machine data,
with the provenance attached.
Managed AI Services applies AI to your machine data, with assurance, an intelligence layer and agents that do work within your boundaries. Every answer traces back to the event underneath it, because without provenance an AI outcome is not evidence. General AI projects outside machine data are not what we do.
AI on machine data, with the provenance attached.
The top layer of our model. Not general AI projects. We do AI on the kind of data we have lived in for years: logs, events, metrics and traces. That is a deliberate boundary. On machine data we know what an outcome means, we can trace it back to its source, and we are willing to base a decision on it.
Why machine data.
Machine data is the one kind of data that accounts for itself: every line has a timestamp, a source and a cause. That makes it suitable for AI in a way free text is not. You can point afterwards at where an outcome came from.
Many AI initiatives start on data whose origin and quality nobody knows. The result looks convincing and cannot be justified. We turn it around: get the data layer right first, then models, and only then agents that are allowed to do something.
Can you explain where the answer came from?
If you cannot, it is not an answer but a guess with a nice chart. Provenance and reproducibility are not bureaucracy: they are the reason you dare to use an outcome, and soon the reason you let an agent act on it.
AI Assurance.
Oversight of what goes into a model and what comes out. The layer you need before anyone in your organisation uses AI for something that matters, and the layer an auditor asks about.
- Provenance and quality of the data used, recorded per source
- Access and governance: who may use which data, and what must never go in
- An audit trail of AI use: which question, which model, which data, which answer
- Drift checks: does the outcome change while the question stayed the same?
- Fits your existing compliance agreements and the EU AI Act
Nobody knows what the model saw .
A model that gives a good answer on the wrong data still gives a wrong answer. Assurance is not about slowing AI down. It is the reason you dare to let it in more widely.
Discuss assuranceWhat it does in practice.
- Spot deviations no threshold would catch
- Reduce a thousand alerts to twenty clusters with a common cause
- Predict when you will hit a capacity limit
- Turn a plain-language question into a search on your own data
- Summarise an outage together with the timeline that led up to it
AI Intelligence Layer.
A layer on top of your platform that reads machine data the way an experienced engineer does: seeing patterns, noticing deviations, making connections a fixed threshold never catches.
We build and run that layer on the platform you already have. No separate product alongside it, no second place your data has to go. Every result stays clickable through to the underlying events. Otherwise it is not insight but an assertion.
And as with dashboards: what delivers nothing gets removed. Including a model that works nicely but that nobody acts on.
Agentic AI.
The step from signalling to doing. Agents that investigate, enrich and prepare on their own, and that only execute within boundaries you set in advance.
- Investigate: on an alert, gather the context itself: what happened around it, has this occurred before, which change preceded it
- Enrich: hand over a filled-in ticket, with a timeline and a first diagnosis, so an engineer starts where the investigation stopped
- Propose: a next step with reasoning, not just a conclusion
- Execute within mandate: only what is explicitly allowed, with a rollback path and an audit trail of every action
We build this up in stages. First investigating and proposing only, and executing once practice shows the proposals were right. That is not caution for its own sake: an agent that takes the wrong action costs you more than the outage it was meant to solve.
An agent never gets more rights than a new colleague.
The same rule as with people: first observe, then propose, then act within a defined mandate. With this difference: an agent writes down everything it did. Which makes it easier to account for in an audit than a person.
Talk about agentic AIWhat would an agent be allowed to investigate first at your organisation?.
Start there, without execution rights. Once the proposals hold up, widen the mandate.