Boat, a Paris-based company fresh from Y Combinator, has begun selling persistent cloud virtual machines at what it claims is a fraction of the cost charged by established competitors—a pitch aimed squarely at developers running fleets of autonomous AI agents that need to stay awake for hours or days at a stretch.
The startup's core offering is straightforward, if audacious: Ubuntu VMs starting at $0.036 per hour, which the company says can deliver cost savings compared to alternatives in the market. Each virtual machine comes with full Linux environment access, SSH connectivity, Docker-in-VM capability, and browser-accessible desktops that stream at 60 frames per second. It's infrastructure tailored for workloads that don't fit neatly into the serverless paradigm—agents that probe databases, scrape websites, or orchestrate multi-step workflows over extended periods.
The company runs its sandboxes on AMD Ryzen 9 9950X hosts distributed across three European regions in Germany, Finland, and France. Every VM receives a public IPv6 or IPv4 address, and the service includes disk-level forking for rapid cloning alongside snapshot functionality. The first ten snapshots come free; additional ones cost $1.70 monthly.
Boat's pricing structure follows a subscription model with four self-service tiers. For $20 a month, developers get 100 concurrent VMs. The next rungs—$100, $500, and $2,000—unlock 300, 1,000, and 2,000 concurrent instances respectively, according to documentation hosted on the company's site. Plan credits convert to compute time by the second, with a $20 plan translating to roughly 555 hours on a 4-vCPU, 8-GB configuration. Deployments beyond those thresholds require direct negotiation with the team.
A seven-day free trial offers 25 compute hours and two concurrent sandboxes, though with automatic shutdown after two hours of inactivity.
Speed Claims and Competitor Pushback
Internal benchmarks published by Boat show its VMs completing 1,000 Node.js build-and-test cycles in 36 seconds, compared to 54 seconds for Daytona, 77 seconds for Vercel Sandbox, 89 seconds for E2B, and 105 seconds for Modal, according to the company's own testing. The company's price comparisons list its 4-vCPU tier at $0.036 hourly against Vercel Sandbox at $0.682 and Modal at $0.760, though the recency of those competitor prices is unclear.
Not everyone in the sandbox ecosystem accepts those numbers without scrutiny. Freestyle, a competing provider, published a blog post claiming that Boat's documentation "exposes a consistency problem"—the service allegedly falls back to older Hetzner cloud VMs when standard capacity runs out, albeit at the same advertised price. Whether that represents a material performance gap or merely operational pragmatism remains a matter of debate among developers evaluating the platforms.
The landscape is crowded. E2B bills itself as "The AI Agent Cloud." Vercel Sandbox emphasizes microVM isolation. Daytona touts sub-100-millisecond startup times. Maritime, another Y Combinator company, promises "each of your users their own VM, forever." Each positions slightly differently, but all are chasing the same emerging market: developers building AI agents that need more than ephemeral compute but less than full cloud instances.
The Team Behind the Service
Anicet Nougaret serves as CEO and Kirill Makarov as CTO of the three-person team. The startup was founded in 2024 and went through Y Combinator's Fall 2026 batch. Boat operates under the legal entity Dedale AI Corp. and rebranded from an earlier incarnation called ASCII Box, with new terms of service taking effect in mid-October.
Customer traction, at least as described in public posts by Nougaret and aggregated metrics from third-party observers, suggests momentum. One mid-October summary by the company claimed Boat had crossed 350 paying users with six-figure annual recurring revenue and month-over-month growth exceeding 160 percent. An earlier update from Nougaret stated that revenue had "more than doubled" following an August launch, with usage climbing 45 percent week-over-week for five consecutive weeks. Customer VMs were collectively running a cumulative 90 days of compute daily at that point, he wrote.
Those figures carry the usual caveats: early-stage growth rates are volatile, and raw user counts reveal little about retention or unit economics.
Developer Experience and Integrations
The platform exposes REST APIs and command-line tools for spinning up, forking, and snapshotting VMs. Built-in support for VS Code and Cursor remote development aims to lower friction for engineers already accustomed to those workflows. Developers can expose VM ports via public HTTPS URLs with token-gated access, and the service streams VM desktops through browsers at 60 frames per second—a feature seemingly designed for debugging visual workflows or running GUI applications.
Boat has integrated harnesses for seven agent frameworks directly into its CLI and API: Codex, Claude, Pi, OpenCode, Prime, Kimi, and Mistral. The platform supports Docker inside each VM, persistent background processes, and even Windows application compatibility through Wine. Third-party adapter projects have begun emerging, including Crabbox, which published provider documentation noting Boat's transition from the ASCII Box brand, and integrations with Harbor for evaluation workflows.
The Broader Context
Sandboxing AI agents presents a distinct set of challenges compared to traditional code isolation, as Axios noted. Agents often need network access to external tools, data sources, and APIs—permissions that complicate security boundaries. Sail Research launched competing "Sailboxes" infrastructure positioning for long-horizon agent workloads. The technical problem is clear enough: serverless functions time out, containers restart unpredictably, and full cloud VMs introduce management overhead that feels excessive for what amounts to running Python scripts with internet access.
Nougaret has written that customers reported "saving $500 to $1,400 per day" compared to prior infrastructure, which would extrapolate to $200,000 to $500,000 annually for larger deployments. The startup argues its flat per-VM pricing model and support for thousands of concurrent instances set it apart from container-based and serverless competitors optimized for shorter-lived workloads.
Whether that value proposition holds as the service scales—and whether competitors respond with their own pricing adjustments—will determine if Boat's aggressive cost positioning is sustainable or simply an artifact of early-stage customer acquisition. For now, the company is betting that developers running AI agents will gravitate toward infrastructure priced and designed explicitly for their use case, rather than retrofitting tools built for an earlier cloud era.
