Skip to main content

Microsandbox

Microsandbox provides an open-source runtime that runs each sandbox as a microVM with its own dedicated kernel, creating hardware-level isolation for AI agents and untrusted code. It supports standard OCI-image workflows, making it a drop-in replacement for Docker for untrusted workloads, with a single SDK and API across Rust, TypeScript/Node, Python, Go, and a CLI. The platform includes destination-bound credential substitution, where real secrets are kept host-side and only released to approved destinations.

Traction 7/10
San Francisco, United States · HQ
Founded 20263300+ followers
  • Developer Tools
  • Cybersecurity
  • Software Only
  • Artificial Intelligence
  • AI Agents
Updated 14 days ago

Funding

Funding rounds are not available yet.

Founders

Founder details are not available yet.

Product

Problem

Traditional container runtimes like Docker share the host kernel, which creates a security risk when running untrusted code, such as AI-generated scripts or third-party applications. A compromise in one container can potentially affect the host system, making it unsafe to run unverified code in shared environments.

Solution

Microsandbox addresses this by providing a runtime that runs each sandbox as a microVM with its own dedicated kernel, creating a hardware-level boundary between the sandbox and the host. This approach allows developers to safely execute untrusted code, including generated code and third-party software, without the risk of a full system compromise. The platform supports a standard OCI-image workflow, making it a drop-in replacement for Docker for untrusted workloads. It offers a single SDK and API, so moving from local development to cloud deployment is a simple configuration change, not a rewrite. The runtime is open-source and Apache 2.0 licensed, allowing teams to inspect the security model and reproduce benchmarks. It can be run locally, in the Microsandbox cloud, or on a customer's own infrastructure via a bring-your-own-cloud option.

Target Audience

Primary customers are development teams and enterprises that need to run untrusted code, such as AI-generated scripts, third-party applications, or multi-tenant workloads, in a secure and isolated environment.

Features

  • Each sandbox runs as a microVM with its own kernel, providing hardware-level isolation from the host system.
  • Supports a standard OCI-image workflow, enabling easy migration from Docker for untrusted workloads.
  • Includes a single SDK and API for Rust, TypeScript/Node, Python, Go, and a CLI, with MCP workflow support.
  • Offers a "bring your own cloud" (BYOC) option for deploying the runtime on a customer's own infrastructure.
  • Provides granular network egress controls, including default blocking of private, host-local, and metadata destinations, with options for allowlists or disabling egress entirely.
  • Features destination-bound credentials, where real secrets are substituted host-side only for approved destinations, preventing credential leakage to untrusted sandboxes.
  • Supports policy outcomes of substitute, pass through, block, or terminate the sandbox for outbound requests with secrets.
This profile is AI-generated and may contain inaccuracies.