How GovAI cut cost per user prompt by 25% while growing 5x

GovAI builds AI-powered tools for government agencies, giving municipal, state, and federal organizations a practical way to deploy artificial intelligence responsibly. Co-founded in January 2023 by Alex Millar, the company launched on Laravel Vapor and scaled quickly.

Today, GovAI processes more than 740,000 AI messages (user prompts) per month across two production regions (US and Canada), roughly five times its volume a year ago.

That growth is what made Vapor's limits visible. As GovAI's AI workloads grew longer and their compliance requirements grew stricter, Millar and Principal Platform Developer Gareth Davis began looking ahead. The question was not whether Vapor would eventually constrain them. It was when.

The move to Laravel Private Cloud was seamless and came with immediate benefits:

• The cost per 1,000 messages dropped roughly 25% (from about $22 on Vapor to $16 on Cloud), while monthly message volume grew about 90% between March and August of 2026.

• The CI pipeline run time on deploys dropped approximately 25% after moving to Cloud.

• The marketing site page load time dropped to roughly 40% of its previous value.

• The team completed the staging migration with approximately two minutes of downtime but without code changes.

Watch GovAI's migration story →

Follow the framework

Before GovAI, Millar co-founded Bonfire, a government procurement platform that went through Y Combinator in 2015. He spent a decade building it and brought a hard-won conviction to his next company about how engineering teams should make technology decisions.

"Your job is to deliver value," Millar said. "And in a startup, your job is to deliver value rapidly before you run out of money."

That conviction became a written DevX manifesto he shares with every engineer on his team. The core principle is what he calls “Follow the Framework”: stay inside Laravel, use its defaults, and resist the temptation to build what the ecosystem already provides.

The lesson came from experience. At Bonfire, Millar made early decisions driven by short-term engineer satisfaction rather than long-term business needs, including microservice experiments that created distributed-systems complexity nobody had signed up to maintain. "We weren't really prioritizing the business needs," he reflected. "The business needs to ship faster."

When Lambda wasn't enough

GovAI launched on Laravel Vapor in 2023 with Corry Flatt, Andrew Wilgar, and Omar Salaymeh. For a bootstrapped co-founder, Vapor's value proposition was obvious. "I can just swipe my credit card, type laravel new, and we're good to go," Millar said.

As the product matured, that simplicity gave way to complexity. GovAI's clients are government agencies with specific security requirements. The team layered in Amazon Comprehend and Macie for personally identifiable information detection, antivirus scanning for files flowing through S3, VPC configurations, and audit logging for SOC 2 compliance.

They added sidecar lambdas to run Playwright for web crawling, keeping that workload separate from their primary Lambda to maintain a clean base image.

The architecture worked, but the team could see a ceiling from a distance: Lambda's hard 15-minute execution cap.

GovAI's queue demands were already unusual for a chat product. Agent loops, document processing, and file transcription ran through SQS with near-real-time expectations. Message volume was already climbing steeply, and the question shifted from "can we stay under 15 minutes?" to "how long until we can't?"

“There's no path around that at all,” Millar said. “As soon as my agents need to run for more than 15 minutes, I'm having to do app-layer coordination to make sure the job shuts itself down in a recoverable state. I didn't want to get backed into that corner.”

At the same time, their clients' expectations grew too, especially around security. Compliance eventually sealed the deal. Laravel Private Cloud's roadmap (FERPA/GLB, ISO 27001, and HIPAA, the latter two now available) signaled the platform was investing in the right direction, and the migration also let GovAI separate US and Canadian workloads into distinct AWS accounts, tightening data segregation.

Growth outpaced the old stack

When the team first walked us through their architecture, they processed roughly 120,000 jobs per day and grew at 8% month over month. Now, in just 30 days, GovAI processes around 7.36 million jobs per month.

The cost curve makes the pattern clear: through late 2025 and into early 2026, GovAI’s AWS spend on the legacy Vapor stack climbed in lockstep with message volume, more than doubling between August 2025 and March 2026 as workloads intensified. Every additional prompt showed up on the bill.

In March 2026, on Vapor, GovAI paid roughly $8,500 to process about 390,000 messages, or roughly $22 per 1,000 messages. Five months later, on Cloud, they're paying around $12,000 to process more than 740,000 messages, roughly $16 per 1,000 messages, as the team worked on right-sizing managed queues. That's roughly 90% more work for 41% more spend, and a rough 25% drop in cost per message.

The rate of infrastructure spend growth didn't just flatten. It reversed. Cloud is doing more work per dollar than Vapor was.

On top of that, the 740,000 figure is also a conservative one. It counts user prompts, GovAI's primary interaction point, but leaves out the growing web crawler and document ingestion pipelines running alongside them. Actual work per dollar has moved further than the headline number suggests.

What came along for the ride

Three improvements arrived without extra effort:

Faster marketing site: After the move, GovAI’s marketing site page load time dropped to roughly 40% of its previous value, a straight infrastructure win, as the frontend was kept the same.

Laravel Cloud WebSockets: GovAI had been using Pusher for WebSocket support. Switching to Cloud’s Laravel Reverb-powered WebSockets required almost nothing. "Pretty much zero effort," Davis said. Reverb also gave the team finer control over concurrency configuration, something Pusher's fixed-allotment model does not provide.

Laravel Nightwatch: Connecting Nightwatch on Vapor required custom setup the team had not gotten around to. On Cloud, it was a couple of environment variables. Monitoring was live before the migration was done.

The scorecard

Both GovAI’s production regions run on Laravel Cloud, on Kubernetes pods that scale with queue depth through Laravel's managed queues. The application code has remained untouched since the migration started, and the team is no longer maintaining a workaround for Lambda's 15-minute ceiling.

The scorecard: deploys are 25% faster, the marketing site is at 40% of its previous load time, and cost per 1,000 messages is down roughly 25% while volume climbed 90%. Plus, GovAI now has a stronger compliance posture through dedicated compute and separate AWS accounts per region that appeals to its growing customer base.

With Vapor behind them, GovAI has room to keep pushing further into state and federal work, and whatever the next 5x looks like.

"Easy, secure, and aligned with the future," Millar said. "That's where we are."

Moving from Vapor to Cloud

GovAI's migration follows a path well-suited to any team running on Vapor today.

• Compare Laravel Vapor and Laravel Cloud →

• Watch GovAI's full migration walkthrough →

• Read the Laravel Vapor to Cloud migration guide →