smol machines
Smol Machines provides a platform for running workloads in fast, hardware-isolated Linux VMs, packaged as portable .smolmachine artifacts that run identically on a developer's laptop, in the hosted smol cloud, or on self-hosted servers. The service supports forking live machines into batches for parallel reinforcement learning rollouts and sandboxing untrusted code with network-off-by-default isolation. Each machine defaults to 4 vCPUs and 8 GiB RAM with elastic memory via virtio balloon.
- Developer Tools
- Software Only
- Artificial Intelligence
- Cybersecurity
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
Developers who need to run untrusted code, train reinforcement learning agents, or deploy consistent infrastructure often face a trade-off between the speed of containers and the strong isolation of traditional VMs. Spinning up conventional VMs is slow and resource-heavy, while containers do not provide a hardware-enforced security boundary against malicious workloads. Additionally, bridging local development and cloud deployment typically requires re-packaging or re-configuring the workload for each environment.
Solution
Smol Machines provides a platform for creating and running hardware-isolated Linux VMs, with workloads packaged into a single `.smolmachine` artifact. This same artifact runs identically on a developer's laptop, in the hosted smol cloud, or on self-hosted servers using the open-source smolvm runtime. The service supports reinforcement learning workloads by enabling users to fork a live machine into a batch of episodes for parallel rollouts. Untrusted code is sandboxed inside a VM boundary with its own guest kernel, with network access off by default and configurable per-machine network policies. The lightweight VMM is linked directly into the binary with no host daemon, and memory usage is elastic via virtio balloon, so the host only commits resources the guest actually uses.
Target Audience
Primary users are software developers and ML engineers building reinforcement learning pipelines, running untrusted code in production, or deploying consistent microVM infrastructure across local and cloud environments.
Features
- Hardware isolation via native hypervisors: macOS Hypervisor, Linux KVM, and Windows Hypervisor Platform
- Portable `.smolmachine` artifact that runs unchanged across local, cloud, and self-hosted environments
- Fork a running machine into a batch of episode machines for RL training rollouts
- Network-off-by-default sandboxing for untrusted code, with configurable network policy and public access per machine
- Elastic memory allocation via virtio balloon, with default provisioning of 4 vCPUs and 8 GiB RAM
- VMM linked into the binary (libkrun + custom libkrunfw kernel) with no host daemon required
- REST API, CLI, SDKs, and a web console for machine lifecycle management
- Export machines as `.smolmachine` artifacts for portability and backup