IAXT is live on Product Hunt today, and it's free: https://www.producthunt.com/prod...
It's a small, free, macOS menu-bar app that passively records what AI coding agents actually do on your Mac: commands, file changes, installs, network calls. Each action is attributed to its session, so you can review what happened after the session ends instead of trying to catch everything live. It never blocks, and everything stays local.
I built it after a long session where I approved what I assumed was a routine scheduled task, and it turned out my agent had set up a login item I didn't notice until later. Harmless in my case, but it made me want a record I could actually review afterward.
Would love feedback from anyone who runs agents locally!
Last month I asked Claude Code to wire up analytics on my Mac. Over one session it ran a dozen shell commands, edited a few files, installed a package, and, as a side effect I never noticed, wrote a macOS LaunchAgent to keep something running at login. I saw none of it live. By then I was clicking Approve out of habit.
Afterward, IAXT’s Action Log showed every action as a row, each attributed to that agent session, with a confidence pill and a severity tier. The LaunchAgent sat near the top in the Review tier: PERSISTENCE: LaunchAgent. Right-click, Copy row as context, paste into another LLM for a second opinion.
That is IAXT: a flight recorder for AI coding agents. It passively records what Claude Code, Cursor, Aider and others do on your Mac, so you can review it after the session. Free, local-first, no account, macOS 13+.
What has your agent done that surprised you?
The right-click copy-as-context gesture is such a smart move — turning passive logs into actionable second-opinion prompts without breaking flow. Nice that it's local-first too.
@pnar1086289 Glad you like it! I have been using IAXT for two months before the release and the copy-paste feature has been useful a bunch of times, especially when some action to review was not clear. I pasted the text as is into ChatGPT and it gave me more context and reassurance on what really happened.
How does it actually capture the agent actions under the hood, like does it hook into shell history or watch filesystem events to piece things together?
@nehir240588 good question. IAXT doesn't use shell history (that only shows what you typed, not what the agent spawns). Instead it watches macOS system events directly:
filesystem events for files created, changed, or deleted
process events for the commands and programs the agent runs
periodic checks of things that don't reliably fire events, like crontab and LaunchAgents, so persistence still gets caught
It then ties each action back to a specific agent session using process ancestry (which process launched what) plus timing, and tags how sure it is: confirmed, likely, or possible. All user-space APIs, no kernel extension, and fully passive, so it never intercepts or slows anything down.
Finally something that shows me what my agents actually did while I wasn't watching. Right-click to copy as context is a really nice touch for second-opinion workflows.
@selim380478 Glad to hear that you like IAXT 😊🚀 I've been using it myself for two months pre-release and I find it really useful. And I do use right-click to copy as context and paste it in ChatGPT or Gemini quite often.