Find out where your tokens go

Every reply in Frontier records what it spent, and so does every tool call inside it. That turns "this is getting expensive" from a feeling into a question with an answer. Start in any thread; see Workspaces & Threads for where the numbers show.

Ask for the last seven days grouped by tool. The answer lists each tool with how often it was called and how much its results added to the conversation, heaviest first, along with the single heaviest calls and the threads they happened in.

Open one of those threads and choose Replay from the reply's menu. Each step shows its own tokens, so you can see the call that read a whole page when a line would have done.

Then change the thing that's actually expensive. Often it's one agent reading too much on every check-in: give it a smaller question, a longer interval, or a cheaper model through the Temperature picker. Ask again next week, grouped by agent this time, to see whether it worked.

Try asking

"Where did my tokens go over the last 7 days? Group by tool and show me the five heaviest calls with their threads."

Calls: get_token_usage

"Now group the same week by agent, and open the thread behind the heaviest call so I can see what it read."

Calls: get_token_usage, get_thread