Dedalus Labs

Dedalus Labs provides persistent Linux virtual machines designed to run AI agents and untrusted code. The service advertises machine startup in under 50 milliseconds, preserved filesystems and memory across sessions, and billing based on active compute rather than idle time.

What it does

Dedalus Machines give an agent a full Linux environment with kernel, memory, filesystem, and root-access isolation at the virtual-machine level. This is intended to provide a stronger boundary for untrusted code than the Docker and V8 isolation models shown on the product site. Machines can sleep and wake while retaining their filesystem and runtime state, avoiding the need to repeatedly pull an image, start a runtime, and install dependencies.

The service is positioned for workloads that need a persistent computer rather than a short-lived sandbox. A command-line installation script is provided, and the site directs users to documentation and a waitlist for access.

Notable capabilities

• Machine startup advertised at under 50 milliseconds, compared on the site with a 2.5-second traditional sandbox flow.
• Persistent filesystem and memory across sessions, with no cold starts claimed by the vendor.
• Full Linux environments with root access and VM-level kernel, memory, and filesystem isolation.
• Active-compute billing: CPU and RAM are billed per second while a machine is running, while idle time is free according to the pricing FAQ.
• Configurable machine sizes and plan limits, including up to 16 vCPUs and 64 GiB of RAM per machine on Pro.

Who it helps

Dedalus Labs is aimed at developers building AI agents that need to execute code, retain working state, or operate in isolated Linux environments. It may also fit teams evaluating infrastructure for agent fleets, especially where persistent environments and stronger isolation for untrusted code are important requirements.

How it fits a workflow

An agent can be assigned a machine, use a full Linux filesystem and runtime, and return to the same environment in a later session without reinstalling dependencies. This model is suited to multi-step coding, automation, or execution workflows where state needs to survive between runs. Teams can select machine resources, keep storage between sleep and wake cycles, and pay for active compute rather than leaving a continuously running machine provisioned.

Strengths and limits

The main stated strengths are fast startup, persistent runtime state, full Linux flexibility, and VM-level isolation. The pricing page lists a Hobby plan at $0 per month with a one-time $20 sign-up credit, up to five machines, and a 50-hour monthly compute ceiling. Pro costs $20 per month and includes $20 of monthly compute credit, while Enterprise pricing is custom and adds options such as SSO, RBAC, audit logs, dedicated fleets, SLAs, and bring-your-own-cloud support. Resource, machine-count, and storage limits vary by plan, so teams should check the official pricing details against their workload before adoption.