earendil-works/gondolin

Health v2 · maintenance only

Adoption not part of the score

2034 stars · 124 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Gondolin is a TypeScript library and CLI that runs untrusted AI-agent code inside fast local Linux micro-VMs (QEMU by default, optional libkrun backend). Network egress and filesystem access are controlled programmatically from the host, with secrets injected only for allowed destinations to prevent exfiltration.

Use cases

• sandbox AI agent generated code before running it
• run curl commands with secrets the guest can never see
• restrict which hosts a sandboxed process can reach
• snapshot and resume isolated dev sessions
• give agents a disposable Linux environment with controlled filesystem access
• prevent credential exfiltration from untrusted code

When to choose

• you run LLM-generated code locally and need egress and secret protection
• you want programmable network/filesystem policy in JavaScript
• you need fast-booting micro-VMs on macOS or Linux without cloud sandboxes

When to avoid

• you need production-grade, audited isolation guarantees
• you are not on macOS or Linux
• you need Windows sandboxing or GPU workloads inside the VM

Facets

library · maturity experimental

security networking secrets-management cli developer-tools security developer-tools self-hosted cli microvm qemu libkrun agent-sandbox egress-control secret-injection typescript ai-agent-safety sandboxing virtualization ai-agents linux macos nodejs

Member repositories

For agents

markdown · JSON · MCP: product_card(name="earendil-works/gondolin")