The Claude Code Control Stack — ccusage + Router (Stop Guessing Tokens)

追加: 2026-05-19

[00:00:00]Claude code feels amazing until you ask one simple question, where did the tokens go and why did every job use the same model?

[00:00:12]Talk repo editors pick two Claude code assets that answer those exact questions.

[00:00:17]>> [music] >> CC usage shows the cost. Claude code router controls the route.

[00:00:21]Mock usage is the moment Claude code stops feeling invisible.

[00:00:25]You can see daily usage, monthly totals, session cost, model breakdowns, and the expensive days you should actually investigate.

[00:00:35]That changes the feeling. Before, you think the workflow was fine.

[00:00:40]After, you know what happened, which session burned the budget, and whether the result was worth it.

[00:00:48]Claude code router fixes the second problem. Not every request deserves the same model. A quick edit is not a deep reasoning task. [music] A background job is not a long context refactor.

[00:01:04]Router lets Claude code send different jobs to different models.

[00:01:08]>> [music] >> Default, background, think, long context, and web search. That is control, not guesswork. [music] Together, they make a loop. Run Claude code, measure it with CC usage, route the next session better.

[00:01:23]Find the Claude code control stack on Talk repo. Install both, then run one real session and measure it.