Why does an AI system need ongoing care at all?

Because its inputs change even when its code doesn't. Providers update and retire models, your data and customers shift, and costs move. Without scheduled evals, quality drops silently.

Can you look after a system you didn't build?

Yes, after a short onboarding review so we know what we're taking responsibility for. If it needs hardening first, we'll say so.

What happens to unused engineering days?

They roll over one month, then expire. It keeps the plan honest for both sides.

After the first three months, yes, with a month's notice. You keep everything — the eval sets, dashboards and documentation are yours.

← ALL SERVICES

AFTER LAUNCH

Monthly · 3-month minimum

Agent Care

Models get deprecated, data drifts, prompts that worked in spring quietly degrade by autumn, and costs creep. Agent Care is the monthly work that keeps an AI system as good as the day it launched — or better.

Ask about Agent Care →€1.5k – €5k / monthtypically €2.5k / month

monitoring
evals
model upgrades
cost tuning
monthly report

WHO THIS IS FOR

Built for you if…

01
Anyone who launched with us and wants the system watched rather than forgotten.
02
Teams running an AI feature or agent workflow built by someone else, with nobody in-house who owns its quality.
03
Companies whose model provider just announced a deprecation and who need someone to migrate safely.

WHAT YOU GET

Everything included.

Monitoring and alerting

Errors, latency, spend and escalation rates watched continuously, with alerts routed to us — not to a dashboard nobody opens.

Scheduled eval runs

The golden dataset re-run every month and after every change, with new real-world cases added as they appear. If quality slips, you hear it from us first.

Model upgrades and migrations

When a better or cheaper model ships, or yours is deprecated, we test it against your evals and switch only when it measurably wins.

Cost tuning

Caching, routing cheap requests to cheap models and trimming prompts. Savings appear in the monthly report as a number, so you can weigh them against the fee.

Agent boundary reviews

Escalation rates and exception queues reviewed monthly. Rules get loosened where agents have earned it and tightened where they haven't.

Small improvements

A set number of engineering days each month for fixes and small features, so the backlog doesn't wait for a new project.

Monthly report

One page: quality, cost, incidents, what we changed and what we recommend next. Written for whoever signs the invoice.

WHAT IT LOOKS LIKE

Some of what you get back.

EVAL RUN — 500 GOLDEN CASESrun 47 · vs run 46

retrieval@5	0.94	+0.02
answer_grounded	0.98	+0.01
refusal_correct	0.91	−0.03
citation_resolves	1.00	0.00
p95_latency_ms	1,840	−260

refusal_correct regressed — build blocked, not shipped. Quality is a number we watch, not a claim we make.

COST & LATENCY — ENFORCED IN CODElast 24h

cost / run€0.11
ceiling €0.50
p50 latency740ms
budget 2s
daily spend€38
cap €80 → queues
escalation rate6.2%
tuned weekly

Budgets are code, not a monthly surprise. Crossing the per-run ceiling switches to a cheaper path; crossing the daily cap queues and alerts.

HOW IT WORKS

The process.

01
01
Onboarding
Week 1
For systems we built, nothing to do. For anyone else's, a short review to set up monitoring, evals and access before we take responsibility.
02
02
Baseline
Month 1
First full eval run, cost and quality baseline, and a list of the quick wins.
03
03
Run and improve
Monthly
Monitoring, eval runs, tuning and small improvements, with the monthly report at the end of each cycle.

TIMELINE & PRICING

What it costs.

DURATION

Monthly · 3-month minimum, then cancel any month

BUDGET RANGE

€1.5k – €5k / month

Typically €2.5k / month

PAYMENT

Monthly in advance.

WHERE IN THE RANGE YOU LAND

·€1.5k/month: one system, monitoring, monthly evals and report, one engineering day.
·€2.5k/month: where most land — model upgrades, cost tuning and up to three engineering days.
·€5k/month: several systems or agent workflows, business-hours response times, and up to six engineering days.
·Systems we didn't build carry a one-off onboarding review, quoted up front.
·Model, hosting and third-party costs stay on your accounts.

EMBEDDED ENGINEER

When you need more than care — a named senior engineer working in your team part-time on an ongoing roadmap — we do that too, on a monthly basis with the same reporting.

STACK

What we build it with.

Observability
Eval harness
Cost dashboards
Your existing stack

NOT INCLUDED

Where the line is.

×24/7 on-call — business-hours response unless agreed separately.
×New products or large features — those are scoped as a build.
×Model, hosting and API spend.

Contracts, ownership & handover →Security & data handling →

FAQ

Questions we get.

Why does an AI system need ongoing care at all?: Because its inputs change even when its code doesn't. Providers update and retire models, your data and customers shift, and costs move. Without scheduled evals, quality drops silently.
Can you look after a system you didn't build?: Yes, after a short onboarding review so we know what we're taking responsibility for. If it needs hardening first, we'll say so.
What happens to unused engineering days?: They roll over one month, then expire. It keeps the plan honest for both sides.
Can we cancel?: After the first three months, yes, with a month's notice. You keep everything — the eval sets, dashboards and documentation are yours.

BUDGET · €1.5k – €5k / month

Have an AI system nobody is watching?

Tell us what it does and who built it. We'll tell you what keeping it healthy would cost each month.

Ask about Agent Care →