Writing
Notes from the boundary.
Long-form pieces about why agent sandboxes look different from CI runners, what containment options exist, and where agentvfs sits in that landscape.
-
Comparing agentvfs to bare Docker for agent code-execution
A side-by-side walkthrough: what it looks like to run an agent's code-execution step on raw Docker, what it looks like on agentvfs, and where the workspace-runtime model earns its keep — plus where Docker still wins.
dockercompareagents -
Designing for agent-aware filesystem access
What a filesystem looks like when its primary user is an LLM, not a human. Vaults, forks, checkpoints, structured output, and the small set of invariants that make an agent loop survive contact with concurrent state.
filesystemagentsruntime -
Execution boundary patterns: namespaces vs MicroVMs vs WASI
A guide to the boundary options for agent code execution. Linux namespaces, MicroVMs, WASI, and the workspace-runtime model agentvfs implements — what each one enforces, where each one costs you, and how they compose.
isolationarchitecturewasi -
Why agents need a different sandbox than CI runners
CI sandboxes optimize for hermetic, single-pass builds. Agent sandboxes need cheap forks, rollback, policy at the command line, and a workspace that survives across turns. They look different — and treating them as the same is where most failures live.
sandboxagentsdesign