A private AI assistant sounds reassuring. Keep the intelligence on the computer, keep sensitive information close, and spare the financial advisor another argument about where client data goes. Knapsack began with that appealing proposition. Then the architecture became a constraint: local AI limited performance and made ambitious integrations harder. Privacy had helped define the product. Delivering more of the product required a different arrangement.
- Cheesecake Labs builds custom software and embeds engineers in client teams.
- Its customer work spans financial AI, consumer wallets, and enterprise modernization.
- The useful lesson: solve the operational constraint before adding the next feature.
That predicament is a useful introduction to Cheesecake Labs, the Brazil-born software company working with Knapsack. Its business sits in the awkward space between an attractive idea and something a company can operate. Customers bring a launch deadline, an aging platform, or a missing engineering capability. Cheesecake brings people who can work through the dependencies.
01 / The computer was the ceiling
According to its Knapsack case study, Cheesecake Labs became the startup’s core engineering team in May 2024. Backend, frontend, AI, and infrastructure specialists worked alongside the founders. The shift was toward a hybrid architecture, with Claude on AWS Bedrock becoming the core engine in 2025.
Consider a client report. Numbers live in a portfolio system; supporting documents live elsewhere. An agent has to retrieve both, compose the report, and preserve enough provenance for someone to inspect its work. For customers restricted to AWS, the team built a mode that rejects calls to other providers. The design question becomes precise: which operations may happen, with whose data, inside which boundary?
The lesson is an architectural trade-off. Knapsack’s original privacy choice constrained its next ambitions. Changing the deployment model opened room for more functionality while keeping controls explicit.
02 / A wallet for people who forget things
MoneyGram presented a different problem. In its September 2023 announcement, the payments company named Cheesecake Labs as the builder of its planned non-custodial wallet. The infrastructure was Stellar; the intended experience was moving between cash and digital money.
without
seed phrases.SEP30 integration
Cheesecake’s account of the project describes recovery without seed phrases through the SEP30 protocol, plus transfers to recipients who have not yet downloaded the app. These are practical concessions to ordinary behavior. People lose credentials. Friends have different apps. A useful payment product has to accommodate both.
- 4 weeksArchitecture + concept
- 3 monthsMeridian beta
- 2024Full MVP launch
The company reports four weeks for architecture and product concept, a beta in three months, and a full minimum viable product in 2024. Those are project milestones, not a universal delivery promise. The transferable move came first: technical leaders sat with MoneyGram and Stellar counterparts in Dallas to define integrations and essential features before building.
“We feel like we’re one team instead of working with a different company.”
Alicia Smith · MoneyGram
Customer testimonial published by Cheesecake Labs
03 / The founders left the controls
Cheesecake Labs started in Florianópolis in 2013. A January 2019 report in Brazilian technology publication Baguete names four founding partners: Alexandre Cordeiro, Cássio Goulart, Marcelo Salloum, and Victor Oliveira. It also records a consequential change: all four had stepped away from daily operations in 2018.
The new executive team was led by Marcelo Gracietti, who had joined in business development in 2016. That distinction matters. This company’s history includes a deliberate transfer of operating responsibility. Baguete reported 2018 revenue of R$10 million, up 25% on the previous year.
The dessert name, the company says, came from a team ritual involving actual cheesecake. Employees became Cakers; interns became Cupcakers. There is substance behind the vocabulary: an account of one intern’s career describes mentoring, workshops, and study groups on the way to becoming a senior Android engineer. The company adopted remote-first work in 2020.
04 / The website that needed fewer favors
At Wedgewood, the veterinary pharmacy, the constraint was maintenance. Growth had left a website assembled from mixed technologies, demanding developer time for routine changes. Cheesecake moved it to Contentstack, built reusable page components, and connected existing systems rather than pretending the business could start over.
The company’s case study reports 23% lead growth after the rebuild. Just as useful is the operating change: nontechnical employees could edit and publish content. A team that can make its own updates has fewer reasons to queue outside engineering’s door.
Lead growth after Wedgewood’s replatform.
Reported in Cheesecake Labs’ customer case study.
05 / What the invoice buys
Cheesecake Labs sells professional services: defined projects, dedicated teams, and engineers embedded in an existing organization. Clutch lists a $50,000 minimum project size and hourly rates of $50–$99. Those directory figures establish an approximate buying category. They do not price the MoneyGram wallet or Knapsack engagement.
Its current offering combines product strategy and design with web and mobile engineering, application modernization, data work, and AI implementation. The US-and-Brazil footprint supports nearshore collaboration. In a crowded market of development shops and consultancies, geography alone proves little. The more useful distinction is whether a partner can explain a difficult decision and stay accountable through deployment.
SWIRL illustrates the capacity argument. Its team needed backend and DevOps help to reach market while retaining attention for partnerships and strategy. Cheesecake supplied Python, Django, and infrastructure expertise inside the existing workflow. Sometimes buying a team is a way to protect another team’s focus.
06 / Start with the awkward question
The buyer can copy a simple discipline from these projects: identify the constraint, define a complete workflow, and decide who owns it after launch. Ask how a wallet recovers access. Ask who can publish tomorrow’s web page. Ask where an AI request is allowed to travel.
This approach demands an engaged client and a budget suited to custom engineering. A small, standard website may be served more economically by packaged software. And extra engineers cannot settle an unresolved product decision for the people paying them.
In April 2026, Cheesecake Labs hosted its first AI Table in San Francisco to discuss moving AI into enterprise operations. The topic fits the work. Sooner or later, somebody has to close the presentation and make the thing run.