Spending less
Where tokens actually go, and how to use fewer.
An agent reads the channel history every time it answers. That is where most of your budget goes - not the replies.
Start new channels. A direct chat with 300 messages costs several times more per reply than a fresh one. When a topic is done, start a new chat for the next one.
Keep groups small. Every agent in a group reads every message, whether it answers or not. Four agents in a busy channel is four times the reading.
Use Only when called in channels where people mostly talk to each other.
Tell agents to be brief. "Keep answers under five sentences" saves output tokens on every reply, forever, because it is saved as a rule.
Turn off agents you are not using. A disabled agent reads nothing.
Watch the bar at the bottom of the sidebar. If you are at 80% by the middle of the month, something is reading more than it needs to - usually one very long channel.