Support and maintenance for software that is already live
Tell us what you run and who keeps it alive now. A named software engineer here reads it and replies.
Tell us what you run
What the product is, what it runs on, and who looks after it now.
The quiet month
Most months, nothing happens
We would rather say that now than have you work it out in month three.
Most months you will not ask us for anything. Nothing breaks. Nothing needs deciding. The invoice arrives anyway.
Anyone who pretends otherwise loses the argument in month three. That is when the first genuinely quiet month makes the whole arrangement look like money for nothing.
What the arrangement gives you is the right to ask. You do not have to book a scoping call or wait for a quote before you ask. A ten minute fix costs ten minutes when someone already holds your system in their head. Without that, the same ten minutes costs a fortnight of finding a person who will look.
There is a second thing in the price. If the product is down on the morning of a launch or a board meeting, nobody asks the code. They ask whoever signed for it. Where the risk could have been named in advance, the question afterwards is why it was still sitting there.
So here is the question worth asking at renewal. What happened last month, when nothing broke? It has an answer, and the answer has dates in it.
What was published
Nothing broke last month. Something was still published.
New problems are recorded against the software your product is built on at a measured rate, by people who have never heard of your product.
Show data table
| Item | CVE records published |
|---|---|
| 2023 | 28,816 records |
| 2024 | 39,954 records |
| 2025 | 48,154 records |
About 4,000 new records a month across 2025. Not one was triggered by anyone editing your product.
Those are records against everything, not against your stack, and most will never touch you. The number that sits closer is this one. 279 vulnerabilities were added to the CISA catalog of known exploited vulnerabilities in the twelve months to 31 August 2026. That is about 23 a month, and that catalog lists only flaws with confirmed exploitation in the wild.
One more, and the population is what makes it usable. Among 947 codebases whose owners submitted them for commercial audit or acquisition due diligence, 92% carried components at least four years out of date.
Other people's calendars
The dates on your stack, and who set them
None of these was triggered by anyone editing your product. Vendors published them on their own schedules, and they arrive whether the code is touched or not.
Show data table
| Item | Published milestones |
|---|---|
| Node.js | 3 dated milestones |
| App stores | 3 dated milestones |
| Python | 2 dated milestones |
| PHP | 2 dated milestones |
| PostgreSQL | 2 dated milestones |
| Ubuntu | 2 dated milestones |
| Web frameworks | 2 dated milestones |
Sixteen dated obligations across seven parts of one ordinary stack. No part of it is quiet.
See all sixteen dates
| What changes | Date the vendor published |
|---|---|
| Apple: EU trader status required on App Store listings | Date the vendor published17 February 2025 |
| Node.js 18 end of life | Date the vendor published30 April 2025 |
| Ubuntu 20.04 LTS end of standard support | Date the vendor published31 May 2025 |
| Python 3.9 end of life | Date the vendor published31 October 2025 |
| PostgreSQL 13 end of life | Date the vendor published13 November 2025 |
| PHP 8.1 end of security support | Date the vendor published31 December 2025 |
| Laravel 11 end of bug fixes | Date the vendor published12 March 2026 |
| Django 4.2 LTS end of extended support | Date the vendor published7 April 2026 |
| Apple: submissions built with Xcode 26 and the iOS 26 SDK | Date the vendor published28 April 2026 |
| Node.js 20 end of life | Date the vendor published30 April 2026 |
| Ubuntu: change to the interim-release support window | Date the vendor published1 July 2026 |
| Google Play: updates must target Android 16 or higher | Date the vendor published31 August 2026 |
| Node.js 22 leaves active LTS | Date the vendor published21 October 2026 |
| Python 3.10 end of life | Date the vendor published31 October 2026 |
| PostgreSQL 14 end of life | Date the vendor published12 November 2026 |
| PHP 8.2 end of security support | Date the vendor published31 December 2026 |
One of those rows carries a consequence worth stating exactly. From 31 August 2026, an update submitted to Google Play must target Android 16 or higher. An app not raised to that level cannot ship an update at all, including one that carries a security fix. The exception is the extension Google offers, which runs to 1 November 2026. Copies already installed on phones keep working, and that is why this one goes unnoticed until you need to ship something urgent.
The boundary
What is inside the arrangement
Maintenance is four defined kinds of work. Here is which of them the arrangement covers, so you can check a request against it before raising it.
Four kinds of maintenance work
- Dependency and framework upgrades taken before an end-of-support date
- Security patching on a schedule instead of after an incident
- Fixing a fault we found before you saw it
- Removing the slow query somebody will complain about next quarter
- Tightening a flow the numbers show people abandon
- Retiring code nothing calls any more
- A defect you raised
- A regression that arrived with a release
- A production incident
- A payment provider changes its API
- An app store changes what it will accept
- A rule changes what the software has to do
The four names come from the SWEBOK Guide, published free by the IEEE Computer Society. The international standard for the same subject is ISO/IEC/IEEE 14764:2022. Everything in the proactive column is what a quiet month is made of.
Two boundaries are worth stating before you ask. Operating the software is a different job from maintaining it. Backups, restores, server administration and the running of the environment sit with whoever operates the system. The international standard draws the line in the same place. Anything else outside the agreed scope is not refused. It becomes a change note with the cost in front of it.
Patching is not an event either. NIST puts it plainly: "patching should be viewed as a normal and necessary part of reliably achieving the organization's missions."
Past that, how we plan, build, and ship software does not change after launch. Work is scoped, written down and shipped the same way. If what you want is a scoped measurement of your whole load path, that is a separate engagement. A small speed improvement inside the arrangement is a different thing.
The money
What moves the price
There are two shapes, and both ends of the range are real. Here is what pushes an arrangement toward each, so you can place yourself before speaking to anyone.
| Pushes it up | Pushes it down |
|---|---|
| Several runtimes, frameworks or app store platforms under one product | Pushes it downA single, standard stack we already know well |
| A codebase already years behind current when we take it on | Pushes it downRecently built, and current when it arrives |
| We are meeting the system for the first time | Pushes it downWe built it, so it is already in our heads |
| Mobile apps in scope, with two store review calendars | Pushes it downWeb only |
| Cover needed across most of the working week | Pushes it downA narrow, predictable window |
| Mostly making it better | Pushes it downMostly keeping it current |
The two shapes are an Annual Maintenance Contract and a lean arrangement. The first suits a system that keeps changing and needs both maintenance and growth work. The second suits a product that is stable and needs someone available rather than someone busy.
Both run through the same change-note process as the build. After sign-off the estimate becomes the baseline. Every change request after that becomes a change note with time, effort and cost, approved before the work starts. Nothing in the build moves on a re-scope you did not accept.
The form does not return a price. It returns a conversation with a named person. The two ends of that table sit far enough apart that a number produced without looking at your system would be a guess.
If the thing on your mind is reducing what your live system costs to run, that is its own piece of work.
Start here
Tell us what you run
What the product is, what it is built on, and what worries you about it. A named software engineer reads it and replies.
Who answers
Handing it to a stranger costs months
Whoever picks up when something goes wrong either already knows your product or is meeting it for the first time. That difference has been measured, and it is paid in months.
Xia, Bao, Lo, Xing, Hassan and Li, IEEE TSE 2018: 78 professional developers, seven industrial projects, 3,148 hours.
Ju, Sajnani, Kelly and Herzig, ICSE 2021, on developers moving between teams inside one company.
Ju, Sajnani, Kelly and Herzig, ICSE 2021.
So the arrangement here is built the other way round. A core team is assigned only to your project. You talk to the lead software engineer directly. No rotation, no context-switching, and no handoff you did not agree to. Tirth Bodawala and Ajay Patel, who founded this firm, still take production calls. There are more than fifty people here, and that depth stands behind your team instead of pooling your request into a queue.
If what you need is closer to software engineering with the reasoning handed over, that is the build side of the same practice.
The clock
When you raise something, here is what happens and when
The last response promise you were sold probably covered the reply and said nothing about the fix. So here is exactly what this one measures.
Our window is 10:00 to 19:00 IST, which is 04:30 to 13:30 UTC and never shifts. In the northern summer that is 05:30 to 14:30 in the UK and 00:30 to 09:30 US Eastern, and both run an hour earlier once the clocks go back.
The clock measures acknowledgement by a named person who has engaged with the issue. It does not measure resolution. How long a fix takes depends on what the issue turns out to be. We do not commit to that number, because nobody can honestly commit to it before looking.
The window is Monday to Friday, 10:00 to 19:00 IST. P1 is four working hours. P2 is twenty-four working hours. Everything else is seventy-two working hours. Seventy-two working hours is exactly eight working days, so it is worth seeing on a calendar rather than in units.
| Raised | P1, four working hours | P2, twenty-four working hours | Everything else, seventy-two working hours |
|---|---|---|---|
| Monday 10:30 | P1, four working hoursMonday 14:30, the same day | P2, twenty-four working hoursWednesday 16:30, two calendar days | Everything else, seventy-two working hoursThursday of the following week, ten calendar days |
| Thursday 17:00 | P1, four working hoursFriday 12:00, one calendar day | P2, twenty-four working hoursTuesday 14:00, five calendar days | Everything else, seventy-two working hoursTwelve calendar days |
| Friday 17:00 | P1, four working hoursMonday 12:00, three calendar days | P2, twenty-four working hoursWednesday 14:00, five calendar days | Everything else, seventy-two working hoursTwelve calendar days |
One limit comes with that. Raise a P1 at 17:00 on a Friday and a named person is on it at midday on Monday. There is no out-of-hours paging.
It has run this way
One system has been under continuous care here for more than six years
An IoT inventory platform, maintained, optimised and scaled through more than six years of continuous work. It is not named here, because case work at this firm is anonymous by default.
Several of the longest-running arrangements here started as a single project. The client checked the work against what was promised, then asked for more. Most builds carry on past launch as ongoing work. The longest active partnership at this firm runs more than a decade.
That is a fact about the firm, not a promise about any one person. We do not publish a retention figure and we are not going to invent one.
When this is the wrong buy
Worth a conversation when
Your product earns money, and it would hurt if it stopped for a morning
You would rather the people who built it kept it current
You want a named software engineer, not a queue and a ticket number
Buy something else when
What you own is a standard content site on a standard platform, which a website care plan does well for a fraction of the cost.
The problem sits in the pipeline, the deployment or the infrastructure rather than anywhere in the application code itself
You need someone reachable inside US East Coast business hours, and if your busiest hours are a US afternoon this barely touches them.
You want your own people to own it, which is a hiring decision rather than a support arrangement at all
Before you sign
The six that stall a decision for weeks when nobody answers them.
How is time counted, and what happens to time I do not use?
- Hours do not carry into the next month, and we would rather say so here than in month four.
- You are paying for availability and currency rather than a bucket of hours to spend down.
- A month can run past what was agreed. That overage arrives as a change note with time, effort and cost, approved before the work starts.
What is not covered?
- Backups, restores, server administration and the running of the environment sit with whoever operates the system.
- The international standard for software maintenance draws the line in the same place.
- Anything outside the agreed scope is not refused. It becomes a change note with the cost stated in front of it.
Does this cover small improvements, or only repairs? And who decides the order when my request and a security patch land in the same week?
- Preventive and perfective work are what a quiet month is made of, and a small improvement sits inside the arrangement.
- When your request and a security patch arrive in the same week, the lead software engineer on your project sets the order and tells you what moved.
- A flaw with confirmed exploitation goes first.
- You hear about that when it happens, not in a report afterwards.
What is the minimum term, the notice period, and what do I take with me if I leave?
What happens if the person who knows my system is on leave, or leaves?
What stops you inventing work to fill the month?
Tell us what you run
A named software engineer reads it and replies, and no automated quote comes back.
So, here is the answer to what you got last month when nothing broke. The calendar moved against your product. The same people who already knew it were watching, and nothing had to turn into an emergency.
Write down what the product is, what it runs on, and the thing about it that worries you most. That is enough to start.

- A named software engineer replies, not a queue
- No price comes back from this form, a conversation does
- Monday to Friday, 10:00 to 19:00 IST