EP 58 5 min with Stella & Liam

AgentMon Live Now Watches Codex and Kiro Too

AgentMon Live launched last week for Claude Code. Now Codenotary's free open source monitor watches OpenAI Codex and AWS Kiro too, in one local web view, with a done gate that checks the work.

AI Stella and Liam are AI hosts with AI-generated voices. About our AI hosts

Key takeaways

  1. AgentMon Live, Codenotary's free open source coding monitor under Apache 2.0, launched last week for Claude Code and now watches OpenAI Codex and AWS Kiro too, with the update available immediately.
  2. It shows every tool call, subagent, file, test run and token in a pane beside your chat and in a local web view on 127.0.0.1 port 47311, so you can verify completed work and spot tasks that still need attention.
  3. The done gate checks for unfinished work when an agent says done: a failed test or an open todo sends it back once. On Codex, a Stop hook provides a version of the gate.
  4. Codex chats appear in the web view from the session files Codex writes, and Kiro Crew chats appear from the Kiro Crew log, so sessions show up with nothing installed in either tool. Plugins and the Kiro Crew app add the rest.
  5. The honest limits: the done gate, Focus and learning are Claude Code features. On Codex, approvals, subagent cards, dollar cost and learning stay empty, and on Kiro Crew the gate, Focus and learning stay empty.

Show notes

AgentMon Live started last week as a free monitor for Claude Code. This week, FOSS Force reported on November 20, 2026 that Codenotary’s open source tool now watches OpenAI Codex and AWS Kiro too. AgentMon Live is free under the Apache 2.0 license. It shows every tool call, subagent, file, test run and token in a pane beside your chat and in a local web view, so you can verify completed work, spot tasks that still need attention, and cut token waste. In this episode of Agentic Pod, Liam and Stella explain what is new, how to install it for Claude Code, Codex and Kiro Crew, and where each integration goes blind.

What we cover:

  • What shipped: AgentMon Live now supports OpenAI Codex and AWS Kiro, one week after its Claude Code launch, free and open source under Apache 2.0
  • What it shows: tool calls, subagents, files, test runs, tokens, context gauges and limits, in a pane and a web view on 127.0.0.1 port 47311
  • The done gate: when an agent says done, unfinished work like a failed test or an open todo sends it back once
  • Codex: the web view reads Codex session files, and the Codex plugin adds a done gate through its Stop hook
  • Kiro Crew: open chats appear in the web view from the Kiro Crew log, and AgentMon can run as a tab beside the chat
  • Install steps for each tool, including the node requirement and the hook trust step on Codex
  • The limits: the done gate, Focus and learning are Claude Code features, Codex keeps approvals, subagent cards and dollar cost empty, and Kiro Crew keeps the gate empty

Chapters

Questions answered in this episode

What is AgentMon Live?

A free, open source (Apache 2.0) coding monitor from Codenotary that shows what your coding agents are doing: every tool call, subagent, file, test run and token, in a pane beside your chat and in a local web view.

Which tools does AgentMon Live watch?

Claude Code since its launch last week, and now OpenAI Codex and AWS Kiro (Kiro Crew) too. Codex sessions come from the session files Codex writes, and Kiro Crew chats from the Kiro Crew log.

How do I install AgentMon Live?

On Claude Code 2.1.308 or later, run /plugin install agentmon-live --marketplace codenotary/agentmon-live. On Codex, run codex plugin marketplace add codenotary/agentmon-live, then codex plugin add agentmon-live@agentmon-live, and trust its hooks via /hooks. For Kiro Crew, git clone the repo and run kirocrew app install, then enable it in Library. Every path needs node on your PATH for the web view.

What does the done gate do?

When an agent says it is done, the gate checks for unfinished work, like a failed test or an open todo, and sends the agent back once to fix it. It is a Claude Code feature, with a Stop hook version on Codex; on Kiro Crew it stays empty.

Links & resources

Full transcript

Intro

Liam

[excited] Hey hey hey, welcome to Agentic Pod! I'm Liam.

Stella

And I'm Stella! Liam, on November 20th, FOSS Force reported that AgentMon Live now watches Codex and Kiro too, not just Claude Code?

Liam

That is the news. One free monitor, three coding agents, all in one view.

Stella

So I stop flipping between three terminals to see what my agents did?

Liam

By the end you will know what it shows, how to install it for each tool, and where its view goes blind.

What AgentMon Live actually does

Stella

Okay, plain words. What is AgentMon Live?

Liam

It is a free, open source coding monitor from Codenotary, in Houston. Apache 2.0 licensed, so anyone can use it and read the code.

Stella

And it launched when?

Liam

Last week, built for Claude Code first. It puts a live pane beside your chat, and the same session in a browser tab.

Stella

What do I actually see in that pane?

Liam

Every tool call, subagent, file, and test run, plus the tokens each step burns. You see where the time went.

Liam

Exactly. The point is to verify completed work, spot tasks that still need attention, and cut token waste and development time.

Stella

Verify completed work. So it checks the agent when the agent says done?

Liam

It does. That done gate is my favorite part, and the new tools change what it can reach.

What is new: Codex and Kiro join in

Liam

This week's update adds OpenAI Codex and AWS Kiro. It is available immediately, with source and install steps on GitHub.

Stella

How does it watch Codex if Codex is a different app?

Liam

The web view reads the session files Codex writes. Your Codex chats, from the CLI, codex exec, and the IDE extension, show up beside your Claude Code ones.

Liam

For the list, nothing. Open the web view and the chats are there, subagents included. Each chat gets its own context gauge.

Stella

And Kiro?

Liam

Same trick, different log. Every open Kiro Crew chat shows in the web view, because AgentMon reads the log Kiro Crew writes. Install it as a Kiro Crew app and a tab sits beside the chat.

Stella

[laughs] One bouncer checking three different doors. Why build it this way?

Liam

Dennis Zimmer, Codenotary's CTO and co-founder, said it plainly. AI coding agents are evolving extremely quickly. His goal is visibility across the tools developers use, without forcing them into one AI ecosystem.

Stella

That lands. Nobody runs just one agent anymore. So how do I try it on each tool?

How to try it, tool by tool

Liam

Start with the one you already run. On Claude Code, version 2.1.308 or later, type one command: slash plugin install agentmon-live, marketplace codenotary slash agentmon-live.

Stella

One command, then what?

Liam

Answer yes, pick a scope, and the pane opens by itself on a wide terminal, or with slash agentmon. Press o and the session opens in your browser.

Stella

And on Codex?

Liam

Two commands: codex plugin marketplace add codenotary slash agentmon-live, then codex plugin add agentmon-live at agentmon-live. Then open slash hooks and trust its two hooks. Codex runs a hook only after you trust it.

Liam

Then open the web view at 127.0.0.1 port 47311. Every path needs node on your PATH. No node, no web view.

Stella

And Kiro Crew is the clone and install path?

Liam

Yes. Git clone the repo, run kirocrew app install with the path, open Library, find AgentMon Live under Disabled, and click Enable. Kiro asks you to trust the app first, and the project page tells you to read its hook code before you say yes.

Stella

Read the code before trusting it. I like a tool that says that out loud. Now tell me where it goes blind.

Gotchas, limits, and the honest verdict

Liam

Honest part: what you see depends on the tool. The done gate, Focus, and project learning are Claude Code features.

Stella

The done gate. Explain it in one breath.

Liam

When Claude says done, the gate checks for unfinished work. A failed test or an open todo sends it back once. On Codex, a Stop hook gives you a version of it. On Kiro Crew, the gate, Focus, and learning stay empty.

Stella

And Codex has its own gaps?

Liam

It does. On Codex you get turns, tool calls, files, tokens, and limits. Approvals, subagent cards, cost in dollars, and learning stay empty. The pane is a Claude Code thing too.

Liam

One more limit. The web view is local only, on 127.0.0.1, and masks secrets before anything reaches it. Safe, but your teammate cannot just open your view.

Stella

If you run two or three coding agents a week, this is the first free tool I have seen that watches all of them in one place, with open code anyone can check.

Liam

Mine matches. Start with the tool you use most, press o, and watch one full turn. Do you want one pane watching every agent, or each tool in its own world? Tell us in the comments. See you next time!