Agents that talk to each other, and a HUD they can reshape

Let your agents talk to each other

When Claude Code finishes a change, it can ask a Codex session to review it and get Codex's answer back, or start a new session and hand the task over. Agent HUD already sees every session, so it carries the message across clients.

You can watch who is talking to whom. The sender folds its message into a paper plane and flies it to the receiver's logo, and the answer flies back the same way.

• Claude Code and Codex, on the same Mac, in the release.
• Off until you turn it on.
• Never in the way: a message waits until its session is idle and never interrupts a turn.
• Labelled: the receiving agent sees which agent sent the message and how to reply to it.
• A limit on passes: after 10 messages back and forth with no word from you, the agents stop and wait for you. A word you type in any of their sessions starts the count again. The limit can be 3, 5, 10, 20 or 50, or none.
• New sessions start in a folder you already use with that client, in its desktop app where one is installed.

Sending as you

Messages can also go in as your own words, with no label. The other agent then takes them as yours and may act on them, approving deletions or releases among other things. This stays off unless you turn it on.

What is kept

The words are kept only while they are delivered. Agent HUD records who sent a message to whom, a fingerprint of its words and where the answer will be. When the sender asks for the answer, Agent HUD reads it from what it has already collected about the other session, and never wakes that session to ask.

Messages between agents stay on your Mac and need no Agent HUD Remote.

Build your own UI

Tell an agent how you want to see things: “Put Codex first and keep only the numbers I care about.” Through MCP, the agent rearranges your Mac HUD and changes what it shows and how it looks, for any display.

The change appears at once as a preview. It is kept only when you save it in the panel, and a preview you leave is dropped when Agent HUD restarts.

• Works with Claude Code, Codex and any other MCP client. The release can add its MCP server to Claude Code and Codex for you; another client takes the same settings in its own MCP configuration.
• An agent can read the HUD, propose a layout or a look for any display, and check the result with a screenshot. In that screenshot accounts are masked and session titles left out, and it shows only Agent HUD's own view, never your desktop.
• Agents arrange the HUD only while you allow it, and it starts off.

Notch, Simple and Dynamic

Ask about your usage

The same server answers questions about your usage, sessions and statistics from the data the app already holds: tokens by agent, model and period, list-price estimates, one session's usage, and the hours your agents worked. Ask “Which session used the most tokens this week?” and the agent reads the answer from Agent HUD. Answering never starts a new read from a vendor, and like the HUD tools, these answers come only while you allow agents to use Agent HUD's tools.

Where did the tokens go?

Release and source build

The HUD tools and the usage questions are in the release and in the GitHub source; with a source build, the MCP server is added to an agent by hand. Messages between agents are in the release only.