Platform

AI monitoring that leads to a decision

AI monitoring in Aicura shows what your AI systems and agents are doing in production, measured against the limits and targets you set. When a figure crosses a limit, an alert goes to a person to decide, and Guardian can respond to the same figures, as far as you allow.

Included in Pro and Enterprise. Compare plans

Know when production stops matching what you approved

A vendor updates the model under a system you assessed last quarter, or an agent's error rate creeps up over a week. Monitoring shows you that as a count, a date or a figure with the facts behind it, never a verdict. It reads the last fourteen days, and an older figure still shows the date it was taken, so stale data reads as stale rather than missing. What a change means stays with the person it reaches.

See which systems have gone quiet

The monitoring overview opens on how many active systems and agents have sent data in the last fourteen days, how many have not, how many tracked figures are reporting and how many alerts are open.

The Coverage tab lists every system and agent with the kinds of data received for it, such as traffic, guardrail activity, billed spend, agent decisions and quality signals, least recent first. A system that has gone quiet rises to the top.

Track the figures that matter for each system

A measure is one figure Aicura tracks, such as error rate, read about one thing. That thing can be an AI system or agent in the register, a component running in your network, your whole AI estate, an alert rule or a training course. A measure about a gateway is never offered against an AI system.

Measures include error rate, average latency, token volume, guardrail block rate, undeclared tool share, decision failure share, observed model mismatch, drift checks reporting drift, security finding failure rate and spend against contract. The same figure over two different periods counts as two measures, each with its own limits.

One page for each system

Each system's monitoring page has tabs for measures, KPIs, traffic, spend, decisions, quality, enforcement, alerts and Guardian, over one time window.

Spend shows contracted, billed and observed side by side, with the billed lines beneath.
Decisions shows each action an agent recorded and flags tools it never declared.
Enforcement shows what the gateway's guardrails did for this system alone.

Per-entry monitoring, Spend tab in Aicura

Alerts that wait on a person

An alert is a measure crossing a limit you set. Nothing is actioned automatically. A person accepts, dismisses or defers it with a note, and every decision stays on the alert as a trail.

Set a limit for everything, or for one named system or component. The more specific limit wins.
Each alert shows the figure that raised it and the limit it crossed.
A decided alert can carry statements from the people who stand behind the decision, which the app calls attestations.

Set your own targets and see whether you hit them

A KPI is a target on a measure, held over a weekly, monthly, quarterly or annual period. One result is written when each period closes, against the target in force for it.

Changing a target cuts a new version that applies from the next period. A closed period is never re-read, and a KPI is disabled rather than removed, so its history stays readable.

When a problem should not wait for a person

Alerts wait on a person. Guardian watches the same figures and, where you allow it, responds itself, from raising it with a person to re-running an assessment, tightening the gateway or quarantining a system. You choose how far it can go for each condition, from five levels.

Off: Not watched.
Tell me: Raise it and stop.
Act with approval: Wait for a named approver before acting.
Act and notify: Act, then say so.
Just do it: Act.

How Guardian works

Spot drift, quality changes and model updates

Drift checks run on an engine inside your own environment, against the datasets you select, and each run reports back as a measure. Quality signals arrive for each system. When a model provider publishes a notice about a model one of your systems uses, which the app calls an advisory, each system records its response.

Per-entry monitoring, Measures tab in Aicura

See how long since each piece of work was done

The Cadence tab lists every piece of governance work for each system, agent or course, such as a risk assessment, an owner review, guardrails, a vendor review or a training course, with when it was last done. Each reads Current, Aging or Stale, worst first, and a row opens the page where the work is done.

See the AI your network reaches, not only what goes through a gateway

When your proxy or firewall logs go to a collector, Aicura reads which AI services your network calls each day and holds them as measures like any other.

AI services in use that no register entry covers, and integrations the register never declared.
Calls to MCP servers, large uploads to AI services, and calls from accounts your vendor records do not name.
Registered systems the network has gone silent on, and how many calls your gateways account for and how many they do not.

Questions

What is AI monitoring?

AI monitoring is watching what AI systems and agents do in production and responding when the figures change. Aicura reads traffic, guardrail activity, spend, agent decisions, quality signals and test runs from components you run in your own network, holds them against limits and targets you set, and raises alerts for a person to decide on.

What is AI drift?

Drift is the distance that opens between how a system behaved when you assessed it and how it behaves now. The data changes, usage changes, or a vendor updates the model underneath you, and the assessment stops describing the system you run. Drift engines running in your own environment report each check, and Aicura tracks figures such as checks reporting drift and checks not completing.

Where does AI monitoring data come from?

From components you run in your own network, which Aicura calls Edge. Gateways report guardrail activity and traffic, collectors send summarised usage figures, billed spend, agent decisions and the AI calls in your proxy or firewall logs, and security and drift engines report their test runs, all through one hub. Nothing appears until a hub and a collector are registered.

Does Aicura act on monitoring alerts without a person?

Alerts never act on their own. A person accepts, dismisses or defers each one. Guardian, a separate part of Aicura, can act, but only where you have told it to. Each condition it watches sits at one of five levels, from Off and Tell me up to Act with approval, Act and notify and Just do it. Guardian only takes actions a person could already take in Aicura, and it can be paused without changing any settings.

Can we track what our AI costs?

Yes. Each AI system's Spend tab in Aicura shows three positions side by side. Contracted is the system's share of the contracts in force, billed is what the vendor reported, and observed is estimated from traffic against priced model versions. Billed lines show the period, vendor, model and amount.