Ideas
How to Reduce Token Usage in AI Agent Workflows
Token usage is not just a billing detail. It changes what a local model can do, how quickly an agent responds, and how much room remains for instructions, tool output, and user context.
The simplest improvement is to stop treating context as a dump. Retrieve fewer documents, split them into smaller chunks, rank those chunks against the actual task, and keep source quotas so one long page does not crowd out everything else.
TinySuite tools are built around that assumption: the best context is usually smaller than the easiest context.