Chainguard launches hardened microVM workspaces for coding agents
Workspaces gives every developer and AI coding agent its own isolated microVM with zero-CVE images and short-lived credentials, launching in beta alongside a CI tool called Checks.
Chainguard announced Workspaces on 14 October at its Assemble London event, a hardened microVM that gives every developer and AI coding agent its own isolated environment. It launches in beta alongside Chainguard Checks, a companion tool that lets agents run CI suites and iterate inside the same loop.
Each Workspace is built only from zero-CVE Chainguard OS Packages and FIPS-compliant toolchains, with dependencies checked against the lockfile at runtime rather than pulled fresh from a registry. The environment never holds a developer’s real credentials: short-lived, encrypted substitutes are injected from outside, and an optional egress policy blocks any outbound connection the team hasn’t approved. Every credential call and network request is logged in a tamper-proof record with a signed attestation. A local working tree stays synced to the remote microVM, so the editor and git workflow don’t change, only where the code actually runs. Cuscal, an Australian fintech infrastructure provider, is an early design partner.
Why it matters: the usual choice for letting a coding agent run unattended has been trusting it with real credentials or slowing it down with approval prompts. Workspaces is a bet that isolating the execution environment, so there’s nothing valuable inside it to steal, is the better trade than babysitting every action an agent takes.