Plimsoll

Plimsoll is a sandbox service that runs untrusted AI code and reports the isolation tier for every execution.

Plimsoll runs untrusted code written by AI agents inside a sandbox and reports the isolation tier on every result. Every request can set a floor, which is the weakest tier the client will accept, and code is not executed if the sandbox falls below that level. It features a trial runner that tests agent programs against hidden scenarios and returns SHA-256 hashes to verify results. A separate program, plimsoll-attest, signs checked results and replays runs. It also allows sandboxed code to call APIs via capability grants without holding credentials directly. Supported frameworks include Vercel AI SDK, Google ADK, Agno, CrewAI, Trigger.dev, and Mastra.

Plimsoll at a glance

Best for: Software engineer
Runs on: Web, API, CLI, self-hosted, open source
Pricing: Free
Released: October 2026
Category: AI coding tool
Website: github.com/plimsollmark/plimsoll

Key features

Reports isolation tier for every run
Enforces minimum isolation floors
Tests code against hidden scenarios
Generates SHA-256 run receipts
Grants scoped API access without credentials
Supports session-based interpreters
Signs and replays runs with plimsoll-attest

Use it for

Running untrusted AI-authored code securely
Testing agent code against hidden test cases
Giving chat agents a secure code execution tool
Allowing sandboxed code to call APIs safely

Alternatives to Plimsoll

OpenSandbox — Provides isolated environments with multi-language SDKs and lifecycle management.
CubeSandbox — Provides sub-60ms cold starts and hardware-level microVM kernel isolation.
AI Sandboxes by Buddy — Stateful virtual machines with persistent disks and public HTTPS endpoints.
Sandcastle — Orchestrates AI coding agents in isolated Docker, Podman, or Vercel sandboxes.
Agent Sandbox — Provides a secure and isolated execution layer for autonomous AI agents.
Daytona — It provisions sandboxes in under 90 milliseconds with full security isolation.
VibeBox — Runs AI agents in a fast virtual machine with cached base images.
Sandkiln — Runs untrusted code in hardware-isolated Firecracker microVMs.
AIO Sandbox — Provides an integrated environment for agents in a single Docker container.
Cleanlab — It automatically detects and blocks incorrect AI responses in real-time.
Agentic CleanSlop — Runs entirely inside AI agents with one pasted line.
Promptfoo — It simulates real users to uncover agent and RAG vulnerabilities.

All Plimsoll alternatives