Nothing to show yet. Head to Settings, turn on Sync with Claude Code, and follow the short setup. Once Claude Code starts running with the helper script, usage data will flow in here automatically.
Tokens in your prompts. Kept as its own chart so the much larger output numbers don't flatten this into nothing.
Tokens Claude wrote back. Usually an order of magnitude bigger than input.
Claude's reported spend in US dollars per bucket. Sums all models and tools.
Distinct Claude Code sessions whose first snapshot fell in this bucket.
Time Claude's servers actually spent answering you — i.e. while a response was being generated.
Everything outside Claude's API: reading responses, reviewing diffs, running tools, typing the next prompt.
Code lines Claude reported as added vs. removed per bucket. Net churn = added − removed.
Share of input tokens served from Anthropic's prompt cache. Higher means cheaper: cache reads cost a fraction of fresh tokens.
Running total of every dollar spent from your earliest snapshot to now.
Running total of all tokens (input + output) consumed.
Average spend per line of code Claude produced in that bucket. Trending down = you're getting more efficient; trending up = prompts getting chattier.
Average tokens (in + out) per line of code added. High = lots of discussion per line; low = tight prompts that shipped code directly.
Average 5-hour and 7-day rate-limit fill at each hour of day in your timezone. Higher columns mean you hit rate limits harder at that time.
Same as above but aggregated by weekday. Use it to spot which days you burn the most budget.
Darker cells = more snapshots posted in that hour × weekday slot. Reveals your working rhythm across the whole history.
Breakdown of sessions by the Claude model that was active when the session last ended.
Sessions grouped by the effort knob Claude Code reports (high / medium / low).
How long sessions last: under 5 minutes, up to 30m, up to 2h, or marathon (2h+).
How full the model's context window is across every snapshot. A tail near 90-100% means you often run close to the limit.
Time between one session starting and the next. Lots of short gaps = batched work; a flat distribution = spread out through the day.
Totals per model across every session: how many sessions, tokens, cost, total session time, and code churn they each account for.
Your five most expensive Claude sessions so far.
Five sessions that consumed the most input + output tokens.
Your longest Claude Code sessions from start to finish.
Sessions that produced the most line additions.