Plimsoll - Self-hosted sandbox for AI agent code with isolation tiers
Show HN: Plimsoll, a self-hosted sandbox for code your AI agent writes

Plimsoll runs untrusted, agent-authored code in a sandbox and reports the isolation tier behind every run—process, container, kernel, or vm. Requests can set a minimum tier; if the sandbox is weaker, nothing runs. It also lets agent code call your API without ever holding your credential, using per-run grants and a broker that attaches secrets outside the sandbox. Built-in exam runners test agent code against hidden scenarios, and every result includes SHA-256 hashes for verification. Integrates with Vercel AI SDK, Google ADK, Agno, CrewAI, Trigger.dev, and Mastra.
Every result says how strong that sandbox was. A request can demand a minimum strength; if the sandbox is weaker, nothing runs.