Runtime policy enforcement for Claude Code, Claude Desktop, Cursor, Codex, OpenClaw — and agents running inside your editor.
Naevik runs below the AI agent. Every action it takes is checked against your policy by the operating system’s own enforcement hooks, before it can take effect.
Claude Code tries to run a command, open a file, or reach out over the network.
Before the action completes, an OS enforcement hook — BPF_LSM on Linux, EndpointSecurity on macOS, or a kernel driver on Windows — evaluates it against your rules.
It runs, it’s blocked before it takes effect, or it’s recorded for audit — exactly as your policy says, and nothing logged that you didn’t ask for.
One local policy governs all your AI agents — and every process each one spawns — across what they run, the files they touch, and where they connect.
These rules cover the AI agent and every process it spawns — child, grandchild, all the way down. Shell out to python3 or bash, and that process is held to the same file and network rules. No escaping the policy by launching a subprocess.
Author a local policy on your own machine — no server, no account, nothing leaves the device.
Claude Code reads and writes your repo, but /Users/alex/.ssh and /Users/alex/.aws stay off-limits — and it can’t write anywhere else on disk.
The AI agent reaches only your test server and the API it needs — the rest of your LAN is blocked, and everything else is logged so you can see where it tried to go.
Add a log rule and review it later — exactly what a freshly-installed skill or MCP plugin reached for behind your back.
Naevik Personal is free. Naevik Team — centrally managed for organizations — is on the way.
On-device policy — no server, no cloud, no license key. For Windows and macOS.
Centrally managed policy and audit for a whole fleet, on a server you own.
Native on each OS — macOS shown here. Click any screenshot to enlarge.