Security and privacy

On your machine

With no account, nothing leaves the machine. Everything Memnox keeps is under ~/.memnox, readable only by you.

File	What it holds
`memnox.db`	The log of every decision, append only
`config.toml`	The mode and how long history is kept
`machine.policies.toml`	Rules that apply in every repository, like never reading `~/.ssh`
`bin/`	The small wrappers that check shell commands
`backup/`	Every agent config setup changed, before it changed it

Arguments are never stored, only a fingerprint of them, because an argument list is where a secret would be.

bash

memnox purge          # drop history older than your retention setting
memnox uninstall      # remove everything setup added

Once you join a team

memnox login is the only step that sends anything. Per action, the machine sends:

Sent	Example
The action and its result	`git.push-force`, `deny`
The rule and its reason	`git-deny`
What it touched	a path, a host or a branch
When, and which session	`2026-09-05 14:03`, `ses_9f21`
A fingerprint of the arguments	never the arguments

Never sent: the arguments, file contents, credential values, what an agent printed, and the machine's signing key. If the cloud cannot be reached, nothing stops: the decision is already made on the machine.

bash

memnox whoami     # which workspace this machine reports to
memnox logout     # stop, and keep the rules already pulled

Where a model is used

Never in a decision. Allow, ask and deny come from your rules alone, so the same action always gets the same answer.

The one place a model is used is in the cloud, to suggest a decision from your team's conversations. A suggestion does nothing until a person approves it.

Reporting a vulnerability

Follow SECURITY.md in the runtime repository rather than opening a public issue.