Show what your agent is doing
An agent that works silently feels broken; one that narrates everything is noise. Oxyn splits the job into layers that all read the same AgentState, so they never disagree.
1. One busy signal
Section titled “1. One busy signal”Put one loader where the agent’s answer will appear: a glyph loader or a status indicator. It settles into a check, a warning or a cross when the run ends, waits or fails.
2. Say what it is doing
Section titled “2. Say what it is doing”Next to it, a verb cycler rotates through a verb pack while the agent thinks and pins the tool’s verb while a tool runs (“Running cargo test”). An elapsed timer counts working time and leaves out the time it waited for you.
3. Show the work
Section titled “3. Show the work”Each tool call is a tool call row with a live timer; a turn’s calls group into a tool stack that folds into “Worked for 31s” when the turn ends. Longer tasks get a todo list or a plan card to review first; orchestrators get a sub-agent tree.
4. Stream the answer
Section titled “4. Stream the answer”Text reveal shows streamed text arriving without feeding tokens to screen readers.
Wiring
Section titled “Wiring”Feed your protocol’s events into an AgentMachine (see Agent state) and pass machine.state() and one shared now to every component. Nothing else needs to coordinate them.