Gyred

Gyred manages AI agent sessions. When you set an agent on a longer task, it rarely just runs and finishes by itself. Every so often it hits a fork it will not cross without a human: "delete this directory?", "which of the two versions do I deploy?", "commit or wait?". The session stops and waits.

The Gyred panel keeps these sessions in one place and shows which is working, which has finished and which is waiting on your decision right now. The desktop version is for running many sessions at once, with density and a keyboard at the desk. The mobile version has one narrower job and deliberately does not pretend to do more.

That job is to unblock what is stuck when you are away from the computer. We built a native Android client for that one moment - not a port of the whole desktop, but a tool for one specific thing.

The whole point of an agent is that it works without you. But the moment it stalls on a decision it will not make on its own, all that autonomy turns into waiting. The agent does not move on until you say "yes" or "no" - and if you happen to be away from the desk, that wait drags on for hours.

This is the paradox that ruins the gain: a tool meant to save time wastes it, standing still and waiting for one tap. The longer the path from the agent's question to your answer, the less its independence is worth.

It is tempting to just move the whole panel to the phone and have "the same, only smaller". That is the wrong answer. Running a dozen sessions at once on a small screen is awkward and nobody works that way - the density that is a strength on the desktop turns into chaos on the phone.

So mobile Gyred and the desktop version are not two copies of the same thing but two tools for two different situations. One is for running the work, the other for unblocking it. The table below shows where that line runs.

The heart of the app is one screen: a list of the sessions waiting on a human, sorted so the most urgent are on top. You do not have to dig through every session to find the three that want something from you - they surface on their own.

For each decision you see the context: what the agent is asking and what it did just before. The point is to answer knowingly, not blind - a "yes" tapped without looking is worse than no answer. Once you know what it is about, you approve, reject or write your own instruction to the session, all from the phone.

The bottom of the screen has four tabs and a fifth for the rest, because more than four on the edge of a phone is a toolbar nobody can hit. Today shows what is waiting, what is running and what the accounts have left. Sessions lists every session with its conversation, tool calls folded to one line. Waiting holds the things to answer straight from the phone, and Todo the open points of the plan. More keeps projects, operations, the week's token spend, the forum and settings.

You can also talk to the assistant by voice, and a home screen widget shows the number of things waiting before you even open the app.

A session waiting on a decision is simply an event the app shows and sends an answer back to. It has a clear state, the text of the question and a set of possible answers - nothing more is needed for the phone to know what to display and what to do with your tap.

That simplicity is deliberate. The less the app has to know about the session itself, the easier it runs fast and the fewer places there are where something can drift between the phone and the server.

Agent autonomy only pays off when the human decision is not the bottleneck. Mobile Gyred removes that bottleneck: a session that used to wait hours for your return to the desk now waits as long as it takes to pull out a phone and read one question.

That is all this moment needs - and, on purpose, nothing more. We did not try to make the phone a second desktop. We made a tool for the one thing that makes sense on a phone, and it does that one thing well.