- vendor
- plimsollmark (GitHub, single author)
- whatItIs
- Open-source sandbox service for running untrusted, agent-authored code that reports the isolation boundary (process, container, kernel or VM tier) each run executed behind, with a SHA-256 run record per run.
- hosting
- self-host-only
- pricingModel
- open-source (Apache-2.0); no price for plimsoll itself, third-party backends (E2B, Docker Cloud Sandboxes) are billed per run
- toolModel
- code-exec
- observability
- SHA-256 run record reporting the isolation tier of each run
- maturity
- preview
- notes
- README has a 'Status, plainly' section: 'Pre-1.0, single author, no external users yet', breaking changes land without a deprecation path, and no third-party security audit has been performed. Maturity 'preview' is inferred from that vendor wording (no explicit alpha/beta/GA label), following the catalogue's armature precedent. Go daemon plus official Python, TypeScript and Go clients; README lists integrations for Vercel AI SDK, Google ADK, Agno, CrewAI, Trigger.dev and Mastra. Distinct from E2B/Modal/Daytona/Blaxel/AWS Lambda MicroVMs: it is a self-hosted broker that can dispatch to E2B or Docker Cloud Sandboxes (and Docker, wasm or NVIDIA OpenShell backends) and reports which of four isolation tiers (process, container, kernel, vm) each run used, rather than operating its own managed sandbox fleet. Launch date not stated in the repo; left null.
- sandboxRuntime
- pluggable by provider: QuickJS on wazero (process), Docker runc (container), Docker gVisor (kernel), E2B Firecracker microVMs or Docker Cloud Sandboxes microVMs (vm); NVIDIA OpenShell also listed
