What is Dedalus Labs?

Dedalus Labs is a Y Combinator-backed company building Dedalus Machines, VMs that boot in 50ms with full Linux, persistent storage, and VM-level isolation. The platform provides production compute infrastructure for AI agents, with per-second billing and scale-to-zero pricing where sleep is free.

What are Dedalus Machines?

A Dedalus Machine (DM) is a full Linux VM that spins up in 50ms. Each machine has its own guest kernel with hardware-enforced VM-level isolation, persistent storage that survives restarts, and per-second billing. When your machine sleeps, you pay nothing. The API is three operations: create, exec, sleep.

How fast do Dedalus Machines boot?

Dedalus Machines boot in 50ms using snapshot restore, not image pull. This is state-of-the-art as of 2026 and orders of magnitude faster than existing providers. The speed comes from a custom virtual machine monitor (VMM) called the Dedalus Hypervisor, built on the Cloud Hypervisor project.

What isolation model does Dedalus use?

Dedalus uses VM-level isolation, the strongest isolation model available. Each machine gets its own dedicated guest kernel. The host CPU enforces boundaries in hardware via privilege ring transitions (VMX root/non-root) and Extended Page Tables (EPT). This is hardware isolation, not containers or user-space kernels.

How does Dedalus Machines pricing work?

You get 50 free hours of compute per month. The Pro tier starts at $20/month with higher rate limits. You are billed per-second for active CPU time only. When your machine sleeps, you pay $0. Storage is persistent and priced separately from compute.

How is Dedalus different from E2B, Daytona, or Fly.io?

Most providers force a tradeoff: fast startup but restricted environments (V8 isolates, Pyodide), or full sandboxes with slow boot times. Dedalus Machines give you both: a full unrestricted Linux environment with 50ms boot, VM-level isolation, persistent storage, and scale-to-zero billing. No imposed session limits, no garbage collection.

What can I run on a Dedalus Machine?

Anything you can run on Linux. Dedalus Machines include systemd, root access, any package manager, any runtime, any binary. GPU support for ML workloads. Nested virtualization for building your own sandboxes. No restrictions on languages, frameworks, or system-level tools.

Does Dedalus support persistent storage?

Yes. Dedalus Machines serve fully POSIX-compliant filesystems with a 100% pass rate on pjdfstest. Memory, volumes, and compute survive restarts. You can access files even while your machine sleeps. Compute and storage are disaggregated, enabling live VM migrations with zero downtime.

What is Dedalus Labs?

Dedalus Labs is a Y Combinator-backed company building Dedalus Machines, VMs that boot in 50ms with full Linux, persistent storage, and VM-level isolation. The platform provides production compute infrastructure for AI agents, with per-second billing and scale-to-zero pricing where sleep is free.

What are Dedalus Machines?

A Dedalus Machine (DM) is a full Linux VM that spins up in 50ms. Each machine has its own guest kernel with hardware-enforced VM-level isolation, persistent storage that survives restarts, and per-second billing. When your machine sleeps, you pay nothing. The API is three operations: create, exec, sleep.

How fast do Dedalus Machines boot?

Dedalus Machines boot in 50ms using snapshot restore, not image pull. This is state-of-the-art as of 2026 and orders of magnitude faster than existing providers. The speed comes from a custom virtual machine monitor (VMM) called the Dedalus Hypervisor, built on the Cloud Hypervisor project.

What isolation model does Dedalus use?

Dedalus uses VM-level isolation, the strongest isolation model available. Each machine gets its own dedicated guest kernel. The host CPU enforces boundaries in hardware via privilege ring transitions (VMX root/non-root) and Extended Page Tables (EPT). This is hardware isolation, not containers or user-space kernels.

How does Dedalus Machines pricing work?

You get 50 free hours of compute per month. The Pro tier starts at $20/month with higher rate limits. You are billed per-second for active CPU time only. When your machine sleeps, you pay $0. Storage is persistent and priced separately from compute.

How is Dedalus different from E2B, Daytona, or Fly.io?

Most providers force a tradeoff: fast startup but restricted environments (V8 isolates, Pyodide), or full sandboxes with slow boot times. Dedalus Machines give you both: a full unrestricted Linux environment with 50ms boot, VM-level isolation, persistent storage, and scale-to-zero billing. No imposed session limits, no garbage collection.

What can I run on a Dedalus Machine?

Anything you can run on Linux. Dedalus Machines include systemd, root access, any package manager, any runtime, any binary. GPU support for ML workloads. Nested virtualization for building your own sandboxes. No restrictions on languages, frameworks, or system-level tools.

Does Dedalus support persistent storage?

Yes. Dedalus Machines serve fully POSIX-compliant filesystems with a 100% pass rate on pjdfstest. Memory, volumes, and compute survive restarts. You can access files even while your machine sleeps. Compute and storage are disaggregated, enabling live VM migrations with zero downtime.

Dedalus Machines

Quick Start

# Install the Dedalus CLI

brew install dedalus-labs/tap/dedalus

# Set your API key

export DEDALUS_API_KEY=<YOUR_API_KEY>

# Create a machine

dedalus machines create --vcpu 1 --memory-mib 1024 --storage-gib 10

# Run a command

dedalus machines exec --machine-id dm-<id> -- whoami

# SSH into the machine

dedalus ssh dm-<id>

Backed by Y CombinatorNewFrom Today to A2A: Crossing the Imagination ChasmFrom Today to A2A

Backed by Y Combinator

Full Linux machines in <50ms

The fastest sandboxes for agents. Persistent runtime. Never sleep. Only pay for active compute.

Get Started Documentation

agent-01

IDLEup 3m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

worker-02

IDLEup 10m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

runner-03

IDLEup 17m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

coder-04

IDLEup 24m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

planner-05

IDLEup 31m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

tool-06

IDLEup 38m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

agent-07

IDLEup 45m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

worker-08

IDLEup 52m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

runner-09

IDLEup 59m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

coder-10

IDLEup 66m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

planner-11

IDLEup 73m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

tool-12

IDLEup 80m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

agent-13

IDLEup 87m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

worker-14

IDLEup 94m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

runner-15

IDLEup 101m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

coder-16

IDLEup 108m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

planner-17

IDLEup 115m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

tool-18

IDLEup 2m

$ dedalus start --gpu

snapshot restored in 12ms

agent ready, listening...

agent-01

IDLE

CPU

20%

MEM

30%

Dedalus Machines

Quick Start

# Install the Dedalus CLI

brew install dedalus-labs/tap/dedalus

# Set your API key

export DEDALUS_API_KEY=<YOUR_API_KEY>

# Create a machine

dedalus machines create --vcpu 1 --memory-mib 1024 --storage-gib 10

# Run a command

dedalus machines exec --machine-id dm-<id> -- whoami

# SSH into the machine

dedalus ssh dm-<id>

Get in touch →Full CLI reference →

Infrastructure

The fastest persistent
sandboxes for agents

Machines that start in <50ms

Full filesystem + memory preserved across sessions
0 cold starts. No waiting

Traditional sandbox2.5s

Dedalus Machines50ms

~50x faster

Full Linux. Maximum flexibility

Install anything: apt, pip, npm, cargo

Other sandboxes trade flexibility for fast startup time. Dedalus Machines give you both.

Runtimes

Python

Node.js

Rust

Java

Ruby

Packages

apt

pip

npm

cargo

brew

System

Root access

GPU / CUDA

Docker

500 GB disk

Workloads

ML training

Web scraping

CI pipelines

Code gen

Sandboxes restrict

xNo root accessxNo system packagesxNo GPU supportxSession limitsxRead-only filesystemxNo Docker

Isolation model

Hardware

Kernel

Agent

Comparison

Dedalus VM

Docker

Kernel isolation

Memory boundary

Filesystem isolation

Root access

Untrusted code safe

FullPartialNone

VM-level isolation. Safe for untrusted code.

Full kernel, memory, and filesystem isolation
Root access inside every sandbox
Stronger boundaries than Docker or V8 isolates

Filesystem and runtime persist across sessions

Only pay for active compute
No idle charges. No hidden fees

Idle compute

Dedalus Machine· Awake

dm-prod-a4f

Freewhen idle

sandbox-aAsleep

$0.095/hr

sandbox-bAsleep

$0.085/hr

sandbox-cAsleep

$0.110/hr

sandbox-dAsleep

$0.072/hr

Other containers keep billing while they sleep. Your DM doesn't.

Preserved while your DM sleeps

/home/machine/*intact

Installed packagesintact

Environment varsintact

Background processesintact

Save on compute

Configure

vCPU4 cores

Memory8 GB

Storage50 GB

Active hrs200 hrs

Idle hrs544 hrs

Monthly Estimate

Dedalus$63.62

Ephemeral sandbox$159.81

60%

less than competitors

$96

/mo saved

Ready to ship?

<50ms away from production

No credit card required.

terminal

$ dedalus machines create --vcpu 2 --memory-mib 4096

Machine dm-7f3a created · status: running · 250ms

$ dedalus ssh dm-7f3a

root@dm-7f3a:~# █

Start with $20 credits Read the Docs

Dedalus Machines

Quick Start

# Install the Dedalus CLI

brew install dedalus-labs/tap/dedalus

# Set your API key

export DEDALUS_API_KEY=<YOUR_API_KEY>

# Create a machine

dedalus machines create --vcpu 1 --memory-mib 1024 --storage-gib 10

# Run a command

dedalus machines exec --machine-id dm-<id> -- whoami

# SSH into the machine

dedalus ssh dm-<id>

Dedalus Machines

Quick Start

# Install the Dedalus CLI

brew install dedalus-labs/tap/dedalus

# Set your API key

export DEDALUS_API_KEY=<YOUR_API_KEY>

# Create a machine

dedalus machines create --vcpu 1 --memory-mib 1024 --storage-gib 10

# Run a command

dedalus machines exec --machine-id dm-<id> -- whoami

# SSH into the machine

dedalus ssh dm-<id>

Full Linux machines in <50ms

Dedalus Machines

The fastest persistentsandboxes for agents

Machines that start in <50ms

Full Linux. Maximum flexibility

VM-level isolation. Safe for untrusted code.

Filesystem and runtime persist across sessions

Save on compute

<50ms away from production

Command Palette

Full Linux machines in <50ms

Dedalus Machines

The fastest persistentsandboxes for agents

Machines that start in <50ms

Full Linux. Maximum flexibility

VM-level isolation. Safe for untrusted code.

Filesystem and runtime persist across sessions

Save on compute

<50ms away from production

Command Palette

Full Linux machines in <50ms

Dedalus Machines

The fastest persistentsandboxes for agents

Machines that start in <50ms

Full Linux. Maximum flexibility

VM-level isolation. Safe for untrusted code.

Filesystem and runtime persist across sessions

Save on compute

<50ms away from production

Full Linux machines in <50ms

Dedalus Machines

The fastest persistentsandboxes for agents

Machines that start in <50ms

Full Linux. Maximum flexibility

VM-level isolation. Safe for untrusted code.

Filesystem and runtime persist across sessions

Save on compute

<50ms away from production

The fastest persistent
sandboxes for agents

The fastest persistent
sandboxes for agents

The fastest persistent
sandboxes for agents

The fastest persistent
sandboxes for agents