Paid audit sandbox for AI coding agents

What to build

A paid, hardened sandbox and audit layer for AI coding agents (Claude Code, Codex, Cursor) aimed at professional developers and small teams: per-project VM isolation, full action/diff logs, network allowlists, one-click revert, and CI integration — shipped as a cross-platform (macOS, Linux, Windows) app or CLI with a team dashboard.

A Show HN tool proved there's appetite for sandboxing and auditing AI coding agents, but it's a free, macOS-only hobby project — leaving room for a hardened, cross-platform, team-oriented paid version with policy enforcement and CI hooks.

Demand

Developers running autonomous coding agents want visibility and containment before letting them touch real repos and credentials, and teams adopting Claude Code/Codex at scale need audit trails and policy controls that a solo hobby tool doesn't provide.

• Hacker NewsNews

  Show HN: I built Airlock, to sandbox your AI agent and know what it's doing — 3 points, 2 comments; author notes 'I didn't like that Claude and' [agents] could act without oversight, prompting the build.

Stack

• macOS Seatbelt / Linux namespaces & cgroups
• Windows Sandbox API
• Hyperlight or Firecracker for VM isolation
• eBPF or syscall tracing for action logging
• Electron or Tauri for a cross-platform app shell
• GitHub Actions / GitLab CI plugin for pipeline integration

Solo + AI difficulty

Entry threshold

Low capital barrier but real technical lift: needs solid macOS virtualization/sandboxing engineering (Virtualization.framework, filesystem diffing, network proxying) and polish beyond a weekend hack; a solo builder with AI assistance could ship a sellable version in a few weeks by building on top of or alongside this open-source tool.

Window

6-12 months