Claude · Claude Code usage menu bar
Claude Code usage, against the clock
Dipstick reads the Claude Code limits you already have — session, weekly all-model, and weekly model-specific — and puts each one next to the time left in its own window, so a 40 percent weekly balance three hours into the week reads as a warning rather than as good news.
Claude · Session Example reading
Usage 71% Window 62% Ahead
Limits Dipstick can show
- Session
- The short rolling window Claude Code reports while you work.
- Weekly · All Models
- The whole-week budget, in the popover’s
Provider · Period · Model poolgrammar. - Weekly · model-specific
- Supplied only by the terminal workflow, so it updates on a slower cadence.
Unique to Claude
Claude is the provider whose weekly window is long enough for the comparison to matter most: seven days is plenty of room to burn a budget on a Tuesday and not notice until Friday.
The session and weekly meters can disagree — a comfortable weekly balance next to an almost-empty session is normal, and Dipstick shows both rather than averaging them into one reassuring number.
Because the scrape has to be interactive it cannot pass --no-session-persistence, so Claude Code writes a session file for each run; the app deletes the *.jsonl files in the one workspace directory it created itself and touches no other Claude Code project.
02 / 06 · Mechanism
How the reading is taken
Dipstick reads the locally authenticated claude executable. Here is exactly what happens on each refresh.
Dipstick launches the locally authenticated claude executable in a private controlled workspace and reads its usage screen. There is no API key, no separate login, and no hosted account anywhere in that path.
A separate local /usage check detects Claude’s own statement that overage credits are currently powering Claude Code, without sending a model prompt.
With separate explicit consent, Dipstick can also connect to Claude Code’s status line. A bounded signed helper receives the documented rate_limits fields on every status-line render and projects only usage percentages and reset times into an owner-only local file, so the session and weekly meters update within seconds while you are working.
The terminal workflow stays as the fallback and remains the only source of the weekly model-specific limit, on a slower cadence.
03 / 06 · Stated here, not buried
Status-line capture is optional, consented, and reversible
If Claude Code already has a custom status line, Dipstick asks before replacing it and restores it when the provider is disabled. Uninstalling runs one deconfigure command that prints provider_claude_integration=deconfigured before you delete anything.
Without capture, everything still works: the terminal workflow is the fallback and is the only source of weekly per-model limits either way. Capture buys you seconds, not features.
04 / 06 · Scope
What is never read or stored
The Claude Code transcript is parsed in memory and never logged or persisted; conversation text, session identifiers, transcript paths, working directories and model metadata are ignored by the status-line helper by construction.
Across every provider, credentials, raw responses, transcripts, account identifiers and local paths are not persisted or logged, and usage readings are not persisted at all unless you turn on local history. The privacy page lists this field by field →
05 / 06 · One product
One app, one download, five providers
This is not a Claude edition. It is the same application and the same download as the other four pages below, and the $2 unlock covers every one of them.
06 / 06 · Start
Set up and start reading
Claude Code installed and signed in. Run `claude` once in a terminal and confirm it answers before enabling the provider.