Move staging first
GovAI moved its staging environment with about two minutes of downtime, and production followed.
LaravelCloud
From servers
A migration story
Same application.
A home that stays awake.
Vapor showed a Laravel team could ship without a server to look after. Laravel Cloud keeps that promise and adds the resident processes a serverless function cannot hold: long jobs, open sockets, a booted application.
Follow the move
The idea in one line Teams moved with no application code changes, and switched more of Laravel on.
The same codebaseVapor · then Cloud
Deploy from git · no servers to patchKept
Scale with demand · down to nothing when idleKept
Long-running jobs · Pro queue workers, no fixed limitSwitched on
Open WebSockets · managed ReverbSwitched on
A booted app · Octane, one toggleSwitched on
Isolation · a cluster of your own, Private CloudSwitched on
Code changes reported: none, by GovAI
Cloud connected · operating
Where Vapor stands
Supported, and closed to new sign-ups.
On 23 September 2026 Laravel stopped taking new Vapor accounts. Existing customers keep running on Vapor indefinitely, kept online, secure and maintained. New projects are pointed at Cloud, which Taylor Otwell describes as built on what Vapor taught him.
Existing customersRuns on, indefinitely
Bug fixes and updates continue. Nobody has to move on a deadline (Laravel Cloud team, 8 October 2026).
New customersStart on Laravel Cloud
Sign-ups had fallen to about five a week, in Taylor's words.
The lineageCloud follows Vapor
Same author, same aim: no servers for the team to run. A different shape underneath.
“An evolution of everything I learned building Vapor.”Taylor Otwell, on X, 23 September 2026 · Monitor 227
What changes, and what does not
The contracts stay.
The runtime moves in.
Queues, broadcasts, caches and disks are framework contracts. On Vapor each one is fulfilled by a function and an AWS service. On Cloud the same contract meets a resident process. The code that calls them does not change.
A request
Vapor runs each request in a Lambda invocation. Cloud runs it on replicas behind a load balancer, and Octane can keep the application booted between requests.
- On Vapor
- One invocation per request, scaled by AWS
- On Cloud
- Flex or Pro replicas, autoscaled; Octane is a toggle
- Your code
- Unchanged
A long job
A Lambda stops at fifteen minutes, and GovAI's AI workloads met that limit. On Cloud, managed queues run on their own workers, scaled by queue depth: Flex workers for jobs up to 90 seconds, Pro workers with no fixed runtime limit.
- On Vapor
- Jobs bounded by Lambda's execution limit
- On Cloud
- Managed queues on Pro workers, or a worker cluster
- Your code
- The same ShouldQueue job
A live update
A function cannot hold a socket open, so realtime on Vapor meant a third-party service. Cloud runs Reverb as a managed resource, generally available since December 2025.
- On Vapor
- A separate hosted realtime service
- On Cloud
- Managed Reverb, enabled in project settings
- Your code
- The same ShouldBroadcast event and Echo client
A quiet night
Vapor's per-invocation billing was a real strength for spiky, quiet workloads. Cloud keeps that property differently: Flex compute sleeps when idle, and the new Flex sizes wake in under 500 milliseconds.
- On Vapor
- Billed per invocation; idle costs little
- On Cloud
- Flex scales to zero (new sizes wake in under 500 ms); MySQL Flex can sleep too
- Your code
- Unchanged
Sources: laravel.com/cloud/docs/compute and /queues (read 8 October 2026); Monitor 991 (GovAI, Lambda limit), 501 (managed Reverb GA, 1 December 2025), 563 (scale to zero, 15 June 2026), 472 (MySQL scale to zero, 20 July 2026).
How teams moved
A short project,
not a rewrite.
GovAI
Leave the data where it is
Trybe kept its external MongoDB Atlas database in place and pointed the new environments at it.
Split traffic by weighted DNS
Trybe sent a share of requests to Cloud and raised it as confidence grew, for a zero-downtime cutover.
Watch both sides in one place
Nightwatch compared the old and new environments in application terms: routes, jobs, queries.
Choose the isolation you need
Trybe, GovAI and Shoptimised went to Private Cloud for a dedicated cluster; PyleSoft moved six apps to Cloud.
Sources: Monitor 991 (GovAI), 990 and 570 (Trybe, story and webinar Q&A), 992 (Shoptimised), 999 (PyleSoft).
Their results, in their figures
Four teams.
Four ledgers.
Each figure below is the customer's own, from their published story. They are not averages and not a promise; they are what these teams measured.
Trybe
5s → 2.5sslowest endpoints, about 500 million requests a month- Compute 17% lower than the same workload on AWS directly
- Zero-downtime cutover by weighted DNS
GovAI
$22 → $16cost per 1,000 messages, while monthly volume grew about 90%- No application code changes
- Deploy CI run time about 25% shorter
Shoptimised
42%lower infrastructure cost- Deploys from over an hour to one or two minutes
- A preview environment per pull request
PyleSoft
50%+lower infrastructure cost, about $5,500 a month- Twelve weeks, thirteen sites, 1.5 million requests a day
- Largest downtime one hour, no data lost
Vapor proved the idea.
Cloud gives it a home.
Three questions for a Vapor team: which of your jobs are shaped around fifteen minutes; which realtime service would you rather not pay separately for; and how much isolation does your next customer ask for.