Boxd

Composable, hardware-isolated Linux virtual machines for developers and AI coding agents, with instant forking.

Cover Block

Publicly reported

Field	Value
Name	Boxd
Tagline	Composable, hardware-isolated Linux virtual machines for developers and AI coding agents, with instant forking [boxd.sh]
Headquarters	Netherlands [Tech.eu, September 2026]
Stage	Pre-seed [Tech.eu, September 2026]
Business model	API / Developer Platform
Industry	Deeptech
Technology	Software (Non-AI)
Geography	Western Europe
Growth profile	Venture Scale
Founding team	Co-Founders (3+) [Tech.eu, September 2026]
Funding label	Pre-seed
Total disclosed	~$2,000,000 [Tech.eu, September 2026]

Summary and Signal

PUBLIC Boxd is a Netherlands-based infrastructure startup building persistent, hardware-isolated Linux virtual machines for developers and AI coding agents, and it merits investor attention because it is trying to replace disposable sandbox and container workflows with stateful machines that can be forked almost instantly [Tech.eu, September 2026] [boxd.sh] [docs.boxd.sh]. The company is still early, but the timing is notable: coding agents are pushing more software work into cloud execution environments, and Boxd is positioning itself as the underlying compute layer rather than as an application on top of that trend [Tech.eu, September 2026] [StartupCafe.ro, October 2026].

The founding team consists of Michiel Voortman, Laurentiu Ciobanu, and Hidde Kehrer, with public reporting and company materials consistently tying them to the product vision and the recent pre-seed round [Tech.eu, September 2026] [6IC, September 2026] [LinkedIn, retrieved 2026]. Laurentiu Ciobanu has described the product as "computers in the cloud" for AI programming agents, which is a useful shorthand for the bet: Boxd is not selling simple code sandboxes, but persistent execution environments that retain state across sessions and support snapshotting and live forking [StartupCafe.ro, October 2026] [boxd.sh] [docs.boxd.sh].

The product claim that stands out is speed with persistence. Boxd says teams and agents can copy running machines, including memory state and active connections, in under 100 milliseconds, while its website emphasizes hibernation, snapshotting, and wake-on-demand behavior for Linux VMs [Tech.eu, September 2026] [BizStack, September 2026] [boxd.sh]. If those capabilities hold up in production, the differentiation rests less on raw compute and more on developer workflow, reproducibility, and the ability to parallelize agent tasks without rebuilding environment state each time [Tech.eu, September 2026] [SaaS Rise, September 2026].

On the financing side, Boxd disclosed a $2 million pre-seed round in September 2026 led by BlueYard Capital, with participation from OVNI Capital, Antler, S20, Script Capital, and angels linked to General Intuition and other technology companies [Tech.eu, September 2026] [6IC, September 2026] [The SaaS News, September 2026]. The commercial model appears to be usage-based infrastructure pricing, with public pricing for vCPU, RAM, and disk consumption plus free starter credits, which is consistent with a developer platform trying to reduce adoption friction at the low end while preserving expansion upside if agent workloads grow [boxd.sh].

What to watch over the next 12 to 18 months is straightforward. Investors will want evidence that Boxd can translate a technically interesting virtualization engine into repeatable developer adoption, named production customers, and workload characteristics that justify its architecture over commodity container-based alternatives, none of which is yet established in the public record [Tech.eu, September 2026] [boxd.sh]. The company has a credible early wedge and a backable investor syndicate, but the next proof point is commercial, not conceptual.

No independent source found -- This section relies on a mix of independent coverage and company materials, with several material product claims still sourced primarily from Boxd's website and documentation.

Taxonomy Snapshot

Axis	Value
Stage	Pre-Seed
Business Model	API / Developer Platform
Industry / Vertical	Deeptech
Technology Type	Software (Non-AI)
Geography	Western Europe
Growth Profile	Venture Scale
Founding Team	Co-Founders (3+)
Funding	Pre-seed, total disclosed about $2,000,000

Company Overview

PUBLIC

Boxd is presenting a fairly specific infrastructure thesis early. The company describes itself as building "composable computers" for developers and agents, and its homepage frames the product as hardware-isolated Linux virtual machines in the cloud rather than disposable sandbox environments [boxd.sh].

Public source coverage identifies Boxd as a Netherlands-based startup and names Michiel Voortman, Laurentiu Ciobanu, and Hidde Kehrer as co-founders [Tech.eu, September 2026]. The available company materials in this record do not establish a founding date or legal entity name, so the cleaner public reading is simply that Boxd is operating from the Netherlands with an infrastructure product aimed at developers and AI coding agents [boxd.sh] [Tech.eu, September 2026].

The first clear public milestone is the company's emergence around its product website and documentation, which outline persistent virtual machines, snapshotting, and related developer workflows [boxd.sh] [docs.boxd.sh]. The next dated milestone is Boxd's $2 million pre-seed round announced in September 2026, led by BlueYard Capital, with participation from OVNI Capital, Antler, S20, Script Capital, and General Intuition, according to Tech.eu [Tech.eu, September 2026].

A founder interview published in October 2026 adds some texture to the positioning without changing the basic chronology. Laurentiu Ciobanu described the product as "computers in the cloud" for AI programming agents, which is consistent with the company's own framing and suggests the team wants to own a deeper infrastructure layer than conventional code-execution sandboxes [StartupCafe.ro, October 2026] [boxd.sh].

One source, partially checked -- Based on company website materials and one independent news report for headquarters, founders, and financing chronology.

The Product and the Stack

MIXED

Boxd is selling a different unit of infrastructure than the container-centric tooling most developers know. Public materials describe persistent, hardware-isolated Linux virtual machines for developers and AI coding agents, with the ability to run code, execute tests, and manage software workflows in parallel [boxd.sh] [Tech.eu, September 2026]. The company website frames these as "composable computers," while press coverage says teams can create copies of running machines, including memory state and active connections, in under 100 milliseconds [boxd.sh] [BizStack, September 2026].

The core technical claim is statefulness plus fast duplication. Boxd says its machines can snapshot, fork, and hibernate when idle, and that they retain running processes, open files, and shell state between sessions [docs.boxd.sh] [boxd.sh]. Tech.eu and 6IC both describe the product around persistent virtual machines and live-forking rather than disposable environments, which is the relevant distinction if the buyer is an engineering team operating coding agents that need reproducible context over time [Tech.eu, September 2026] [6IC, September 2026]. Laurentiu Ciobanu reinforced that positioning in an October 2026 interview, describing the product as "computers in the cloud" for AI programming agents rather than conventional sandboxes [StartupCafe.ro, October 2026].

The public evidence is thinner on underlying architecture than on product behavior, so the safest read is to focus on what is externally claimed and priced. Boxd says machines provide CPU, memory, storage, and networking isolation, and its pricing page lists credit-based usage charges of €0.039 per vCPU-hour while running, €0.012 per GiB-hour for RAM, and €0.00008 per GiB-hour for disk, plus €30 in starting credits [docs.boxd.sh] [boxd.sh]. Those details are useful because they suggest a usage-based developer platform with economics tied to active compute and memory, while the no-idle-time positioning depends on hibernation and persistence working as advertised [boxd.sh] [SaaS Rise, September 2026].

No independent source found -- This section relies materially on company website and documentation claims, with partial corroboration from Tech.eu, 6IC, BizStack, SaaS Rise, and StartupCafe.ro.

The Market They Are Entering

Publicly reported The market matters now because the bottleneck has shifted from generating code to giving developers and coding agents reliable places to run it, test it, and keep state between sessions [Tech.eu, September 2026] [StartupCafe.ro, October 2026].

Boxd sits at the intersection of several larger infrastructure categories, but the available public record does not provide a named third-party TAM, SAM, or SOM specific to persistent virtual machines for coding agents. The more defensible framing is analogous rather than direct: coverage consistently places Boxd in cloud infrastructure for developers and AI coding agents, with a product centered on persistent, hardware-isolated Linux virtual machines rather than disposable container sessions [Tech.eu, September 2026] [6IC, September 2026] [The SaaS News, September 2026]. That suggests Boxd is pursuing spend that today is distributed across developer environments, CI runners, ephemeral preview infrastructure, and secure execution environments, then widened by the newer budget line emerging around agentic software development [boxd.sh] [Tech.eu, September 2026].

Demand drivers are visible even in the narrow source base. The repeated product claims point to a user need for reproducible, isolated compute that can run code, execute tests, and manage software-development workflows in parallel, which is a practical requirement once a team moves from one developer workstation to many concurrent software agents [Tech.eu, September 2026] [6IC, September 2026]. Laurentiu Ciobanu's description of Boxd as "computers in the cloud" for AI programming agents is useful here because it implies a broader workload than simple sandboxing, including persistent processes and longer-lived state that many short-lived container setups handle awkwardly [StartupCafe.ro, October 2026].

The adjacent markets are easier to define than the exact core market. Boxd can be read as partly adjacent to cloud IDE backends, CI/CD infrastructure, preview environments, secure sandboxes, and virtual desktop or workstation infrastructure for engineers, because its own materials cite development workspaces, coding agents, preview environments, and CI runners as target use cases [boxd.sh] [docs.boxd.sh]. It also competes with the operational habits teams already have, namely local development on laptops and stateless containerized execution in the cloud, which is why the company's public messaging contrasts persistent machines with conventional ephemeral environments [boxd.sh] [Today’s Startup News, September 2026].

Regulatory and macro forces are mostly indirect, but they still matter. Hardware isolation and workload separation are easier to justify in an environment where enterprises are asking sharper questions about code provenance, access boundaries, and how autonomous agents execute actions inside development workflows, even if the current sources stop short of naming formal compliance regimes or enterprise contracts [Tech.eu, September 2026] [6IC, September 2026]. At the same time, the macro tailwind is straightforward: if coding agents become more common, the infrastructure layer that lets them run many isolated tasks in parallel should grow with them, but that remains a conditional rather than an observed market outcome in Boxd's public record [Tech.eu, September 2026] [StartupCafe.ro, October 2026].

Market lens	Publicly supported claim	Source
Core market	Persistent virtual machines for developers and AI coding agents	[Tech.eu, September 2026]
Workflow adjacency	Development workspaces, preview environments, and CI runners are cited use cases	[boxd.sh] [docs.boxd.sh]
Substitute baseline	Short-lived container environments and local development workflows appear to be the operational alternative	[Today’s Startup News, September 2026] [boxd.sh]
Demand signal	Live-forking and persistent machine state are repeatedly presented as the key differentiators	[6IC, September 2026] [BizStack, September 2026]

The table shows a market that is easier to describe by workflow pain than by a clean analyst category. That is often how new infrastructure wedges form, but it also means investors should treat category sizing as provisional until third-party market research catches up with the product shape.

One source, partially checked -- Section relies on one independent lead publication and several smaller outlets, with company materials used for use-case framing; no named third-party TAM/SAM/SOM report was available in the cited public sources.

The Competitive Field

Positioning

MIXED Boxd is positioning itself against disposable developer sandboxes and container-first execution layers by arguing that coding agents and developers need persistent, hardware-isolated Linux machines that can be copied while live, rather than short-lived environments that reset state between tasks [boxd.sh] [Tech.eu, September 2026] [6IC, September 2026].

Company	Positioning	Stage / Funding	Notable Differentiator	Source
Boxd	Persistent, hardware-isolated Linux virtual machines for developers and AI coding agents	Pre-seed, $2M announced September 2026	Live-forking of running machines, including memory state and active connections, reported in under 100 milliseconds	[Tech.eu, September 2026] [6IC, September 2026] [boxd.sh]
E2B	Cloud runtime infrastructure for AI agents and developers	Named competitor in structured facts	Public materials in the provided research identify it as a competitor, but do not support a more specific feature comparison here	[Structured facts]
sprites.dev dev sh]

The competitive map is narrow today, even if the broader market is crowded. On one side are incumbents and substitutes, meaning local development, conventional cloud virtual machines, CI runners, and container-based preview or sandbox environments. On the other are newer agent-infrastructure challengers such as E2B and sprites.dev, both relevant because Boxd is not selling generic compute, it is selling a specific workflow primitive for stateful software execution by humans and coding agents [Tech.eu, September 2026] [StartupCafe.ro, October 2026] [boxd.sh].

That distinction matters because Boxd's product claims rest on persistence and copy speed, not on a proprietary model layer or a horizontal developer suite. The company says machines can retain running processes, open files, and shell state between sessions, while public coverage also describes snapshotting and live-forking as the core behavior users are buying [boxd.sh] [docs.boxd.sh] [Tech.eu, September 2026]. In practical terms, the immediate substitutes are not only direct startups, but also any stack where an engineering team can approximate the same workflow with containers, ephemeral VMs, or internal tooling.

Boxd's edge, if the public claims hold up in production, is product architecture rather than distribution. The reported ability to create copies of live machines, including memory state and active connections, in under 100 milliseconds is a concrete claim and sharper than the usual "agent infrastructure" framing because it points to a workflow advantage that could matter in testing, debugging, and parallel task execution [6IC, September 2026] [BizStack, September 2026]. The durability of that edge is less clear. It could compound if the company builds a developer ecosystem around APIs, pricing, and repeatable agent workflows, but on current public evidence it still looks perishable because larger cloud or tooling platforms could imitate portions of the workflow if customer demand becomes obvious [Tech.eu, September 2026] [boxd.sh/pricing].

The clearest exposure is that Boxd does not appear to own a distribution channel yet. No named customers, deployment partners, or ecosystem integrations are confirmed in the public record provided here, which leaves room for competitors with stronger developer reach to narrow any technical gap [Tech.eu, September 2026]. E2B is the more obvious threat if buyers decide that agent-runtime convenience matters more than VM-level fidelity, while sprites.dev matters if persistent agent compute becomes a recognizable category and customers accept a narrower implementation with simpler adoption. Boxd is also exposed to adjacent incumbents if teams conclude that existing cloud and CI infrastructure is good enough once wrapped with internal automation.

The most plausible 18-month scenario is a sorting process around workflow fit rather than a single category winner. Boxd is the likely winner if enterprise and startup engineering teams decide that reproducible, stateful machines are the right abstraction for coding agents, especially where live debugging and branching from in-flight state save meaningful engineering time [StartupCafe.ro, October 2026] [Tech.eu, September 2026]. The likely loser if that thesis weakens is sprites.dev, because narrower persistence claims are harder to defend if customers either move upmarket toward fuller virtual-machine semantics or downmarket toward simpler ephemeral runtimes. Boxd itself would be pressured if E2B or incumbent cloud tooling makes the higher-level developer experience easier to adopt, even with less architectural purity.

One source, partially checked -- Relies on Tech.eu, 6IC, BizStack, StartupCafe.ro, and company materials for Boxd; competitor detail is limited in the provided public evidence, and the E2B and sprites.dev rows are constrained by named-competitor references rather than full independent profiles.

Opportunity

PUBLIC The prize here is not another developer tool, it is a chance to become the control plane for persistent compute environments used by both human engineers and AI coding agents, a layer that could sit underneath a meaningful share of software creation if the shift from disposable sandboxes to long-lived machine state proves durable [Tech.eu, September 2026] [boxd.sh].

The headline opportunity is straightforward. If Boxd can make persistent, hardware-isolated Linux machines the default runtime for coding agents, preview environments, CI runners, and developer workspaces, it could occupy infrastructure spend that today is fragmented across local machines, cloud VMs, containers, and bespoke internal tooling [boxd.sh] [Tech.eu, September 2026]. That outcome is reachable, not merely rhetorical, because the public evidence points to a product with a clear technical point of view: persistent virtual machines, snapshotting, and live-forking, including reported copying of a running machine with memory state and active connections in under 100 milliseconds [BizStack, September 2026] [6IC, September 2026]. The company also has an early investor set that fits the thesis for infrastructure software, with BlueYard Capital leading a $2 million pre-seed round and participation from OVNI Capital, Antler, S20, Script Capital, and angels from General Intuition, which suggests Boxd has enough backing to keep building the virtualization engine before broader commercial proof is visible [Tech.eu, September 2026] [The SaaS News, September 2026].

A large outcome can arrive through more than one path, but each path depends on the same underlying bet: persistent machine state matters more as software development becomes more agentic.

Scenario	What happens	Catalyst	Why it's plausible
Agent runtime standard	Boxd becomes the default infrastructure layer for AI coding agents that need isolated, stateful environments to run code, test changes, and preserve working context across sessions	A category shift toward coding agents that need more than short-lived containers, paired with Boxd proving its live-forking and persistence in production workflows [StartupCafe.ro, October 2026] [Tech.eu, September 2026]	Public sources consistently frame the product around cloud computers for AI programming agents rather than disposable sandboxes, which indicates a specific design target rather than a general-purpose VM product [StartupCafe.ro, October 2026] [6IC, September 2026]
Developer workflow substrate	Boxd wins first with human developers, preview environments, and CI runners, then expands into agent use cases from inside existing engineering workflows	Product-led adoption driven by credit-based pricing and use cases already named on the company site, such as workspaces, preview environments, and CI runners [boxd.sh] [boxd.sh]	The platform is already described publicly as supporting development workspaces, coding agents, preview environments, and CI runners, which broadens the entry point beyond a single AI-native buyer [boxd.sh] [Tech.eu, September 2026]
Persistent compute API	Boxd becomes an embedded API for software vendors building agent features into their own products, so Boxd monetizes the infrastructure layer rather than only end-user seats	Launch of stable developer primitives around machine creation, snapshotting, hibernation, and forking that third parties can integrate directly [docs.boxd.sh] [boxd.sh]	The company is presented as an API and developer platform with composable machines, and the product language already centers on programmable infrastructure rather than a closed application [docs.boxd.sh] [boxd.sh]

The compounding mechanism, if it appears, is operational rather than social. A developer or agent that can resume prior state, fork a live machine instantly, and hibernate when idle should reduce setup time, duplicated environment work, and idle compute waste, which makes the product more attractive as workload volume grows [boxd.sh] [boxd.sh]. If those workflows become embedded into CI, preview, and agent loops, Boxd can gain stickier usage because machine templates, snapshots, and team processes become harder to unwind than a lightweight sandbox dependency. The public evidence is early, but it is directionally consistent: the company markets persistent state, snapshotting, hibernation, and live-forking as core primitives, not side features, and that is the right shape for a product that hopes to accumulate workflow lock-in over time [docs.boxd.sh] [boxd.sh] [Tech.eu, September 2026].

The size of the win is best framed through infrastructure ownership rather than near-term revenue. Public sources here do not provide a market size figure or a directly disclosed comparable valuation, so the upside has to be expressed conditionally. If the "agent runtime standard" scenario plays out, Boxd could become the kind of foundational developer-infrastructure asset that strategic acquirers or late-stage investors value on platform importance rather than point-solution utility, because it would control where stateful agent work actually runs [Tech.eu, September 2026] [StartupCafe.ro, October 2026]. In that scenario, not a forecast, the company could plausibly support venture-scale outcomes well beyond a niche tooling exit, since the workload in question spans both human development and machine-generated software operations. The public record is too thin to anchor a tighter valuation range responsibly, but the upside case is still legible: own the runtime for agentic software development, and the addressable spend moves from a small tool budget toward a broader slice of cloud and engineering infrastructure.

One source, partially checked -- This section relies on one strong independent news source, additional smaller publications, and company materials for product capability claims.

Sources