Skip to content
Civware

AI that keeps workingafter launch day

Once your AI is in daily use, it needs care: model updates, cost control, fixes and improvements. A monthly support plan covers that work, with a short report on what changed and why.

Group
Run
First step
A support plan for one system
Price
Monthly, quoted per system
You decide
What gets improved next

When it fits

For systems we built, or ones your team built and would like help running.

  1. 01

    An AI system is live, and nobody has time to watch it or improve it.

  2. 02

    Your model provider is changing or retiring the model your system uses.

  3. 03

    Costs are rising, and you'd like to know why before the next bill.

What we build

  1. 01Monitoring of quality, cost and usage, with alerts to a named owner.
  2. 02Model updates, tested against your examples before release.
  3. 03Fixes and small improvements, released on a regular cycle.
  4. 04Cost reviews, with options to bring spending down.
  5. 05A monthly report in plain words: what changed, what was fixed and what's next.

How it works

Each month follows the same cycle, and nothing goes live without being tested first.

  1. 01Monitor quality, cost and usage.
  2. 02Review with your team and agree what to improve.
  3. 03Every change, including a new model, is tested on your examples before release.
The monthly support cycleEach month runs the same cycle: monitor quality and cost, review with your team, improve with fixes and updates, then release once tested. A monthly report shows quality passed on your examples, costs within the agreed limit, and a model update tested and ready to release.FIG. 01THE MONTHLY CYCLEMONITORquality and costREVIEWwith your teamIMPROVEfixes and updatesRELEASEtested, then liveEVERY MONTHMONTHLY REPORTQuality: passed on your examplesCosts: within the agreed limitModel update: tested, ready to release
Fig. 01 The monthly cycle. Every change, including a model update, is tested before it's released.

Safeguards

Most problems after launch come from change: new models, new data, new costs.

A model update changing results
New models are tested against your evaluation set first, and released only if the results hold.
Costs creeping up
Spending is checked against an agreed limit, with an alert well before it is reached.
Fixes that break something else
Every release reruns the full set of test cases, not just the one that was fixed.
Being locked in
You keep the code, prompts and documentation, so you can move support in house or elsewhere.

How to start

First engagement · Support plan

Monthly care for one system

The plan includes:

  • Monitoring and alerts for quality, cost and usage
  • Tested model updates and fixes
  • A monthly report and review with your team
  • A clear route for raising problems
Price
Monthly, quoted per system
Scope
Agreed for each system
Decision pointEach month, you decide what gets improved next.

Questions

Can you support a system you didn't build?

Often, yes. We start with a short review and an evaluation set, so we know how it performs today.

What happens when a model is retired?

We test a replacement against your examples, show you the results, and switch only once you've agreed.

How is the price set?

Per system, based on how much it is used, how many systems it connects to and how much risk it carries.

Can we stop?

Yes. Everything is documented, so your team or another provider can take over.

Tell us one job you'd like AI to do

We'll tell you honestly whether AI is the right fit, and what a sensible first step looks like.

One job is enough to start.

We use your details to reply to you, as described in our privacy policy.