run it locally

There is no hosted dashboard. That is the point.

AI Monitoring is self-hosted. There is no account to create and no server of ours between you and your data, because the entire premise is that your AI usage never leaves infrastructure you control. Personal mode is the fastest way to see that for yourself: one install, one command, your own usage on your own machine.

$ pipx install aimonitoring-security
$ aim personal
→ open http://127.0.0.1:8787

Requires Python 3.11 or newer. Standard library only — the CLI has no runtime dependencies.

Do not run pipx install aim. That name belongs to AimStack's unrelated ML experiment tracker. The distribution here is aimonitoring-security; the command it installs is still aim.

What it reads, locally

Claude Code session and usage records
Cursor local state
Kilo Code, Kimi Code, and Grok Build usage
Which tools ran, when, how often, and what they cost

What it never does

Make an outbound network call — run it with networking off
Read or store your prompts
Read or store your code or diffs
Ask for an account, a key, or an email address

verify it yourself

Do not take the privacy claim on trust.

The canonical event schema rejects prompt text, response bodies, tool-call arguments and plaintext identities, and a guard script proves it in CI. That script has a self-test mode that deliberately removes each control to confirm the check fails when it should — so the test is itself tested. Both run on every commit, and you can run them without standing up a stack.

Read the guarantees

the full stack

Evaluating it for a team?

Personal mode is a single machine. The full deployment adds the ingest service, Postgres, object storage, the analyst console, shadow-AI discovery and PR security. It runs on a laptop from one script, on loopback, with demo data.

See the deployment paths →