Dedalus Labs review

Dedalus Labs is a managed cloud service that provisions isolated Linux virtual machines, called Dedalus Machines. It sits below an agent framework in the stack: the agent uses the REST API, CLI or SSH to create and control a machine, then runs software and keeps files there between sessions.

Each machine has its own kernel, root access and persistent root filesystem. Files and packages in locations such as /home, /root, /etc, /usr/local and /var survive sleep and wake cycles. Machines move through running, sleeping, starting and destroyed states. Autosleep starts after five minutes of inactivity by default; an API command or SSH connection wakes a sleeping machine. Dedalus says machines boot with zero cold starts. CPU and RAM can be resized through an API call without restarting. This targets agents that need persistent state, long-running tasks or interactive tools, rather than stateless function execution. The compute platform is managed by Dedalus; it is not a library or a self-hosted server.

The main integration surface is an HTTP REST API. Dedalus also provides the dedalus-cli command-line interface and official SDKs for Python, TypeScript, Go, C#, Java, Kotlin and PHP. The Executions API dispatches non-interactive commands and lets clients poll results; SSH provides a pseudo-terminal for interactive tools such as Claude Code, vim and htop. The service runs Linux guest VMs and exposes endpoints such as dcs.dedaluslabs.ai.

The MCP gateway connects models to local or Dedalus-hosted MCP servers, with tool chaining, streaming and model handoffs. Dedalus describes it as vendor-agnostic; it does not publish a numeric integration count. Its client tools, SDKs, CLI and helper libraries are open source on GitHub. Confirm current API behavior, SDK coverage and operational guidance in the vendor's documentation before choosing a client. Observability integrations include PostHog, according to the vendor.

Hobby costs $0 per month and includes a one-time $20 sign-up credit. Each machine is capped at 4 vCPU and 16 GiB RAM, with up to five machines and 10 GiB storage per machine; storage cannot be expanded. Pro costs $20 per month and includes $20 in monthly compute credits, which do not roll over. It allows up to 20 machines, 16 vCPU and 64 GiB RAM per machine, with 20 GiB storage per machine and unlimited expansion. No annual option is listed.

After included credits, active compute costs $0.04536 per vCPU-hour and memory costs $0.01458 per GiB-hour; both are billed per second while the machine runs. Additional storage costs $0.0001 per GiB-hour. Sleeping machines do not incur active CPU and RAM charges, but storage is billed by time. Additional purchased credits carry over; their price is not published. Qualifying startups can receive up to $10,000 in compute credits, valid for 12 months. Enterprise pricing is custom and includes a dedicated fleet option, SSO, RBAC, audit logs and an SLA with uptime guarantees.

Dedalus open-sources its client tools, SDKs, CLI and helper libraries on GitHub. The vendor does not publish a license in the published product details, so confirm the applicable license and its conditions before redistributing or embedding those components. The core VM platform and compute infrastructure are proprietary, managed cloud services. Dedalus documents no self-hosting option.

Linux gives workloads a familiar execution environment, but machine provisioning, autosleep, resizing and execution rely on Dedalus's service and API. Moving away would mean replacing those control-plane integrations and arranging new compute for the workloads. The vendor does not publish a data-export process. SSH provides a way to interact with machine files, but ask Dedalus about bulk export, backups and deletion before storing data that must be portable or retained.

The plan limits are concrete: Hobby allows five machines, while Pro allows 20, with per-machine CPU and RAM caps. Enterprise offers custom limits and an SLA with uptime guarantees; the standard plans do not list an uptime commitment. The vendor also does not publish API rate limits in the published product details. Ask about rate limits, availability, incident response, backups, data retention and security controls before making the service a production dependency.

Machines have isolated kernels and root access, but those facts alone do not establish a compliance posture or a security boundary for every untrusted workload. Confirm the controls and certifications your deployment requires. Autosleep can reduce active-compute charges, but verify how it interacts with background jobs and connection handling.