# Dedalus Labs - Full Reference

> VMs that boot in 50ms. Full Linux, VM-level isolation, persistent storage, scale-to-zero. The production compute platform for AI agents.

## Company Overview

Dedalus Labs builds Dedalus Machines, a cloud computing primitive that is both performant and flexible. A Dedalus Machine (DM) is a full Linux VM that spins up in 50ms.

The company was founded in 2025 and is headquartered in San Francisco, CA. Dedalus Labs raised an $11M seed round led by Kindred Ventures and Saga Ventures, and is backed by Y Combinator (S25 batch).

## Articles

Practical guides to AI agent infrastructure are indexed at https://www.dedaluslabs.ai/articles. Individual article pages remain available at https://www.dedaluslabs.ai/blog/{slug}.

## Dedalus Machines — Core Product

### What is a Dedalus Machine?

A Dedalus Machine (DM) is a full Linux environment. Each DM is a VM that spins up in 50ms via snapshot restore. You are only billed for active CPU time. When your machine sleeps, you pay nothing.

A DM is a full, unrestricted Linux environment. Run any language, any framework, any system-level tool. You don't need to be a DevOps expert. Create a VM, execute in it, tear it down. Three operations.

DMs are persistent by default. Memory, volumes, and compute survive restarts. No arbitrary timeouts. No garbage collection killing your long-running sessions. Your agent can run for hours, days, or indefinitely.

### Speed

Dedalus Machines start up in 50ms, a state-of-the-art number as of 2026 and orders of magnitude faster than existing providers.

This is achieved by building a custom virtual machine monitor (VMM) called the Dedalus Hypervisor, built on top of the Cloud Hypervisor project. The hypervisors are orchestrated in their own Kubernetes pods to reactively and proactively meet the demands of dynamic workloads.

### Security — VM-Level Isolation

There are three types of isolation models, ordered by increasing strength:

1. Container-based isolation (weakest — trust boundary is software, cgroups/namespaces)
2. User-space kernel isolation (medium — software Sentry intercepts syscalls, large attack surface)
3. VM-level isolation (strongest — hardware-enforced via CPU privilege rings)

Dedalus Machines enforce VM-level isolation for maximum security. Each user gets their own virtual machine with its own dedicated guest kernel orchestrated by the Dedalus Hypervisor.

The host CPU enforces the boundary in hardware via privilege ring transitions (VMX root mode for the host, VMX non-root mode for the guest). Guest memory is isolated via Extended Page Tables (EPT) / Nested Page Tables (NPT), which are walked by the CPU's MMU hardware, not by software.

Unlike user-space kernel models (e.g. gVisor), VM-level isolation does not add syscall interception latency and cannot be bypassed by a single software bug.

### Disaggregated Compute and Storage

Compute and storage layers are separated entirely. This flexibility enables:
- Live virtual machine migrations without noticeable downtime
- Independent scaling of RAM and disk
- Access to files even when the machine is asleep

### Durable Storage

Dedalus Machines serve fully POSIX-compliant filesystems, achieving a 100% pass rate on pjdfstest (a common benchmark used by popular open-source filesystems). Currently supports ext4, with planned support for xfs, btrfs, and openzfs.

### Flexibility

Full Linux environment including systemd. Unlike V8 Isolates or WebAssembly, a Dedalus Machine is truly an unrestricted sandbox. Any language, any framework, any binary. Root access, GPU support for ML workloads, and nested virtualization.

## API

Using Dedalus Machines:

```
$ dedalus machines create --vcpu 1 --memory-mib 1024 --storage-gib 1
$ dedalus machines:executions create --machine-id dm-<id> --command '["whoami"]'
$ dedalus machines sleep --machine-id dm-<id>
```

The interface is designed to be native for AI agents via the Dedalus CLI (https://github.com/dedalus-labs/dedalus-cli).

## Pricing

### Dedalus Machines
- **Hobby**: $0/month, $20 one-time sign-up credit, no credit card required
- **Pro**: $20/month with $20 in monthly compute credits and priority support
- **Enterprise**: Custom pricing, dedicated support, SLAs, custom limits
- Per-second billing for active CPU time. $0 when sleeping. Storage priced separately.

## Frequently Asked Questions

**Q: What are Dedalus Machines?**
A: VMs that boot in 50ms with full Linux, persistent storage, and VM-level isolation. Each machine is a real Linux box with its own guest kernel. The API is three operations: create, exec, sleep.

**Q: How fast do Dedalus Machines boot?**
A: 50ms via snapshot restore using the custom Dedalus Hypervisor built on Cloud Hypervisor. State-of-the-art as of 2026. Orders of magnitude faster than existing providers.

**Q: What isolation model does Dedalus use?**
A: VM-level isolation — the strongest available. Hardware-enforced via EPT/NPT and privilege ring transitions (VMX root/non-root). Each machine gets its own dedicated guest kernel. Not containers, not user-space kernels (gVisor).

**Q: How does pricing work?**
A: Every account receives a one-time $20 sign-up credit. Pro costs $20/month and includes $20 in monthly compute credits. CPU and RAM are billed per second while active. Storage is billed separately.

**Q: What can I run on a Dedalus Machine?**
A: Anything that runs on Linux. Full systemd support, root access, any package manager, any runtime, GPU support, nested virtualization. No restrictions on languages, frameworks, or binaries.

**Q: Does storage persist across sleep/wake?**
A: Yes. POSIX-compliant filesystem (100% pjdfstest pass rate). Memory, volumes, and compute survive restarts. Files accessible even while machine sleeps. Compute and storage are disaggregated.

**Q: How is Dedalus different from E2B, Daytona, or Fly.io?**
A: Most providers force a tradeoff: fast startup but restricted environments (V8 isolates, Pyodide), or full sandboxes with slow boot. Dedalus gives you both: full unrestricted Linux with 50ms boot, VM-level isolation, persistent storage, and scale-to-zero billing.

**Q: What is the Dedalus Hypervisor?**
A: A custom virtual machine monitor (VMM) built on the Cloud Hypervisor project. It enables 50ms boot via snapshot restore and is orchestrated in Kubernetes pods for elastic scaling.

**Q: Does Dedalus support live VM migration?**
A: Yes. Compute and storage are disaggregated, enabling live VM migrations with zero downtime during fleet maintenance. Workloads are never interrupted.

## Technical Architecture

- **VMM**: Custom Dedalus Hypervisor built on Cloud Hypervisor
- **Isolation**: VM-level via EPT/NPT hardware enforcement
- **Boot**: Snapshot restore (not image pull) in 50ms
- **Storage**: Disaggregated from compute, POSIX-compliant ext4
- **Orchestration**: Kubernetes pods for hypervisor management
- **Networking**: Full Linux networking stack
- **Compute**: Per-second billing, scale-to-zero
- **Migration**: Live VM migration with zero downtime

## Contact

- Website: https://www.dedaluslabs.ai
- Documentation: https://docs.dedaluslabs.ai/dcs
- Email: support@dedaluslabs.ai
- GitHub: https://github.com/dedalus-labs
- CLI: https://github.com/dedalus-labs/dedalus-cli
- X/Twitter: https://x.com/dedaluslabs
- LinkedIn: https://www.linkedin.com/company/dedalus-labs
- Discord: https://discord.gg/RuDhZKnq5R
- YouTube: https://www.youtube.com/@DedalusLabs
- Y Combinator: https://www.ycombinator.com/companies/dedalus-labs