Reset Radar

Pacing Coach · runs on your machine

A warning light tells you to stop. An instrument tells you how fast.

Claude runs on three clocks: the rolling five-hour window, the weekly all-model cap, and the separate Fable cap on Max plans. The Pacing Coach reads all three off what your machine already knows, and turns them into one answer · which clock binds first, and what is safe to send right now.

Install it What leaves your machine

No account access. No credentials. No API key. Nothing is sent anywhere unless you turn on the away relay, which is off until you do.

What you actually see

Three surfaces, one reading

The engine is the same in all three. Where you look is up to how you work.

In your statusline

real output

One short line, rendered on every turn. It says the binding clock and nothing else, because a statusline that needs reading is a statusline you stop seeing.

40% · 1h43m

Captured from this machine. Forty percent of the five-hour window left, rolling in an hour and forty-three.

Or in a spare pane

The same reading, refreshing, for a second terminal you glance at rather than read.

rr-coach watch --every 20

Costs nothing to leave running: every refresh is a local file read, and the calendar is cached.

In the browser

real output

An extension popup for the work that happens in a tab. It reads from a local endpoint you start yourself, so the extension never talks to us · and it computes nothing of its own, which is why it can never disagree with the terminal.

The Pacing Coach extension popup: room for about 7 more jobs before this resets; five-hour window 76 percent left resetting in 2h50m; weekly cap 72 percent left resetting in 4 days; plenty of room; and the next three dated changes from the record.

The popup, rendering this machine's real Claude Code history through the same engine the terminal uses. Load it unpacked from the directory rr-coach install prints.

The full report

real output

When you want the picture rather than the glance: both windows with their bars, what the record says is coming, and one line of advice in work language rather than percentages.

Reset Radar · Pacing Coach

  Next goodwill flush projected for the first week of August · our
  projection, in about 3 days. Not an announcement.

  Five-hour window     ############ 100% left   resets in 5 hours
                       0 turns · room for ~8 more jobs
                       baseline learned from your own 20 completed windows

  Weekly cap           ------------   ?  left   resets in 4 days
                       160 turns · no baseline yet · needs a few more windows

  > Plenty of room. No reason to think about limits right now.

  From the record
    Aug 2   Next goodwill flush projected for the first week of August
    Aug 5   Doubled Cowork usage limits scheduled to end
    Aug 19  Claude Code +50% weekly limits scheduled to end

rr-coach report. The calendar block is the same record this site publishes, read locally from a cached copy so the report still works offline.

Every tool, side by side

real output

Claude Code and Codex measured against your own history with each, so a tool you use lightly is not flattered by a tool you lean on.

  Tool          Window              Left        Resets in   Turns   Pace
  ---------------------------------------------------------------------
  Claude Code   Five-hour window    ###..  40%   1h44m       143     steady
                Weekly cap          ####.  74%   4d          455     light

  Codex         Rolling five hours  .....   0%   2h45m       24      hot
                Weekly cap          .....   0%   4d          24      hot

rr-coach compare, run on this machine. Two tools, four clocks, one screen.

Two commands

Install

Node 18 or newer. Nothing else.

  1. Install the package

    npm install -g https://resetradar.com/dist/rr-coach-0.1.0.tgz

    Served from this origin, checksummed at —. The registry name @resetradar/coach is reserved for the same build.

  2. See it work

    rr-coach report

    No wiring needed for this. It reads your existing session history and tells you where you stand.

  3. Put it in your statusline

    rr-coach install --write

    Backs up settings.json first, and refuses to replace a statusline you already configured. Run rr-coach install with no flag to see the edit instead of making it.

  4. Staying current

    rr-coach update

    This moves quickly, so the coach checks once a day and mentions a new version on the report · never on the statusline, and never by updating itself behind your back. A tool that reads your session files should not be able to silently replace its own code. Turn the check off entirely with rr-coach config set updateCheck false.

Everything else it can do
rr-coach statusline
the one-line reading, for a statusline command
rr-coach watch
a refreshing report, for a second pane
rr-coach compare
every detected tool, side by side
rr-coach json
the whole assessment, machine-readable
rr-coach providers
what this machine has that can be read
rr-coach serve
the local endpoint the browser extension reads
rr-coach relay --dry-run
print the exact bytes the away relay would send, before enabling it
rr-coach update
upgrade in place · works whichever way you installed
rr-coach license add KEY
apply a licence key

Local-first, and specific about it

What leaves your machine

By default: nothing

Every reading is computed locally from files your tools already write. The coach has no account, asks for no credentials, and works with the network off. The published calendar is cached on first use so even that is not a live dependency.

If you turn on the away relay

One small aggregate, when a session ends: which clock binds, a fraction, and a position in time. That is the whole payload, and rr-coach relay --dry-run prints it so you can read it before you ever send one.

Never, in any mode

  • your prompts or Claude's replies
  • file names, paths, or repository names
  • code, diffs, or command output
  • your Anthropic account or its credentials

The readings come from token counts and timestamps in your own session files. The coach never reads message content, and there is no mode in which turning something on makes it start.

What it costs

The reading is free. Being told while you are away is not.

The instrument stays free permanently. What you pay for is the part that needs a server: reaching you when you are not at the desk.

Free

Nothing, permanently

  • Statusline, full report, and watch
  • How much room is left, and when it resets
  • Every tool you use, side by side
  • The Reset Radar calendar built in
  • The browser extension
Install

No account. Nothing to cancel.

Pro price not final

$6 / month or $60 / year

  • Everything in Free
  • A note when a window fills up while you are away
  • A note when you have been going too long without a break
  • History across machines
Get Pro

Aggregate counts only. Never your prompts, files or paths.

When it does not look right

Troubleshooting

The weekly cap says "?" instead of a percentage

That is the honest answer, not a bug. Remaining quota is not readable from your machine · nothing local knows it. The coach estimates headroom from your own completed windows, and it will not guess until it has seen a few. Keep working; the number arrives once there is something to base it on.

The statusline is blank

Deliberate. A coach that breaks your statusline is worse than no coach, so the hot path fails silently. Run rr-coach statusline in a terminal to see the real error, and rr-coach providers to check that something readable was found.

It says my tool is not present

The reader looks for the session history your tool writes. If you have only just installed that tool, there may be nothing to read yet. rr-coach report --provider synthetic runs the whole thing on demo data, which separates "nothing to read" from "something is broken".

Are these Anthropic's official numbers?

No, and the coach says so on every screen. Published limits are not exposed to your machine. What you are seeing is measured against your own history, in an index of our own · useful for pacing, not a quota readout. Your own /usage remains the authority.

Does it work with Codex, or other tools?

Codex is read today and marked experimental. The reader is a seam rather than a special case: a tool that writes its own session history can be added without touching the engine, which is also what makes the side-by-side comparison honest.