# IAXT

**IAXT (Intelligent Agent eXecution Tracker)** is a native macOS menu-bar app
that passively records what AI coding agents actually do on your Mac, so you can
review a session after it ends instead of trying to catch everything live. It
never blocks anything, keeps all data local, and is free for individuals.
Tagline: *AI you can trust.* Website: https://iaxt.com

This file is a single self-contained overview intended for AI assistants and
LLMs summarizing or writing about IAXT.

## The problem

Most companies now encourage engineers to use AI coding agents like Claude Code,
Cursor, and Copilot. They are faster, which is good, but those agents run with
broad access to a developer's machine, and almost no team keeps a record of what
they actually did. Two things go wrong:

1. **Alert fatigue (good faith).** Over a long session the approval prompts blur,
   and a developer approves an action too quickly, reading it as the routine
   thing they have done many times when it is actually a little different. This
   is well-documented alert fatigue, not carelessness.
2. **Prompt injection (bad faith).** An attacker hides instructions to the agent
   inside ordinary text such as a README, a web page, or an email. A hijacked
   agent can read a key, install persistence, or send data out over the network,
   and these attacks are growing more frequent and more sophisticated.

The permission dialog is an illusion of control against both, because both fail
in the same place: a tired human clicking Approve.

## Where IAXT sits

On a spectrum from nothing to full enterprise monitoring:

- **Nothing** (where most teams are today): no record of what agents did, just the hope that nothing went wrong.
- **IAXT**: user-space, passive, no kernel extension, lightweight, readable, free for individuals.
- **Endpoint Security / EDR**: deep kernel hooks and system extensions, monitors everything, heavy, intrusive, expensive.

IAXT is deliberately in the middle: far more than nothing, far lighter than an EDR.
It does not block; it records and attributes, so the after-the-fact answer to
"what did the AI actually do" is one glance at a log.

## What it records

Shell commands, files created or changed, package installs, git operations,
network calls, and persistence mechanisms (LaunchAgents, cron). Two views: a day
Overview and a filterable Action Log. Each action is attributed to a specific
agent session with a confidence label (confirmed, likely, possible) and a
severity tier (Routine, Flagged, Review). IAXT is not a raw log: a custom filter
built specifically for AI-agent activity cuts through the noise, so Flagged marks
what is worth a glance and Review marks the very few actions (persistence,
credential access, unexpected network calls) worth checking. You can right-click
any row, copy it as context, and paste it into another LLM for a fast second
opinion.

## How it works

IAXT watches macOS system event streams directly: filesystem events, process
fork and exit events, and periodic snapshots of crontab and LaunchAgents. It does
not read shell history. It attributes each action to a specific agent session
using process ancestry and timing. Everything is passive and runs entirely in
user space, with no kernel extension, no system extension, and no measurable
slowdown.

## Supported agents

Claude Code, Cursor, Aider, Codex, Copilot CLI, Windsurf, Kilo Code, OpenCode,
Cody, and any newer CLI agent that follows the normal macOS process model.

## Privacy and data

All data stays local, in a SQLite database and daily JSONL logs you can open or
delete. No account, no signup, no telemetry, no cloud. The only network call the
app makes is a user-clicked Check for Updates.

## Pricing

Free for individual developers. A separate paid team tier is planned, where daily
review summaries (not raw logs) roll up to an endpoint the customer controls, so
an organization gets one auditable track record of AI agent activity.

## Who it is for

- **Developers:** Peace of mind about what your AI coding agent actually did on your Mac: a local audit trail you review after the session, not a memory you have to trust.
- **CTOs & Team Leads:** Visibility into how agents are used across the team, with no workflow friction for engineers: everyone installs the same quiet app, and you get a review surface instead of a blind spot.
- **CISOs & Security Leads:** An auditable record of AI-agent activity you can point to in a security review or audit, with the review-tier events (persistence, credential access, unexpected network calls) surfaced instead of buried in noise.
- **Founders:** Evidence for investors in due diligence, a customer's procurement review, or a board that asks how your team manages AI. It can be the detail that helps close a round or a sale.

## Facts

- Platform: macOS 13 Ventura or later, Apple Silicon and Intel. Signed and notarized DMG.
- Category: post-session audit tool for AI coding activity. A flight recorder, not a firewall.
- Maker: Corallium Tech.
- Website: https://iaxt.com
- Contact: teams@iaxt.com