Pay for what fits.

Decadra reads the session logs Claude Code and Codex write locally, prices your usage under every plan the vendor sells, and reports each provider on its own. Nothing leaves your machine.

Install View source

Free for noncommercial use Runs locally Zero network calls Reads your own logs

claude code, ! shell

!decadra now

heldMax 20x $200.00/mobaseline 2026-08-01

this session$1.27 list-equivalent42 min

today$18.40 list-equivalent

month to date$136.660.7x plan price

pace$421.90 by month end at this rate12 of 30 days observed

ceilingno recorded utilization

last verdicthold, medium confidence2026-08-15

real field names and shape from the source. Dollar figures are a plausible example, not live data.

works with

Why it exists

Token counts measure consumption, not output, and the units differ between vendors. So Decadra never sums or ranks them. It gives every provider its own block, and one opinion about that provider's own plans.

Runs inside the agent's shell

Run !decadra now from the ! shell in Claude Code or Codex. It prints one block in about 200 milliseconds: this session so far, today, and month to date against the plan you hold. No browser tab, no lost context.

Where your usage sits against every tier

decadra assess prices your list-equivalent usage under every plan the vendor sells, so you can see whether a flat-rate tier is comfortably covering you or barely keeping up.

Each provider, its own math

Claude Code tokens and Codex tokens aren't the same unit, so Decadra never adds them up or ranks them. Each gets its own block. The only number that spans providers is what you paid, not what you used.

Local reads, offline by default

Token and cost math run through ccusage, always --offline. Decadra makes zero network calls today. An opt-in price-refresh command is planned for a later release.

what each tool's own usage view already does, checked against public docs, Sept 2026. This changes; verify against the vendor.
tool	native usage view	from the agent shell	prices it against other plans	no separate login
Claude Code	`/usage`: quota bars, burn rate, reset time	yes	no	yes
Codex	`/status` or a web dashboard: quota %, reset time	yes	no	yes
Cursor	web dashboard; a balance figure in editor settings	no	no	no
Devin	web dashboard: ACU balance per task	no	no	no
Muse Code	not publicly confirmed as of this writing	unconfirmed	no	no
Decadra	`decadra now`: one command, Claude Code and Codex today	yes	yes	yes

Get it running

$ git clone https://github.com/rishabbalak/decadra

$ cd decadra && npm install && npm run build

$ node dist/cli.js doctor

Node.js 22.13 or later
Today: build from source (left). v0.4.0 is unreleased, not yet on npm.
Once published: npm install -g decadra gives you the decadra command directly
First run writes ~/.decadra/plans.json with plan prices as of 2026-09-05, hand maintained, never overwritten

core commands
command	what it does
`decadra now`	this session, today, and month to date against the plan you hold
`decadra baseline add`	record what you pay for today, append only
`decadra assess`	effective cost under every plan the vendor sells, one provider at a time
`decadra sessions`	turns, duration, and commit signals per session, medians per provider
`decadra doctor`	checks config, local data, pricing coverage, git, and ccusage
`decadra alerts check`	plan-change alerts from stored verdict history, quiet unless one fires

provider support
provider	commands that produce data
Claude Code	nowassesssessionsdoctor
Codex	nowassesssessionsdoctor
Cursor	nowassesssessionsdoctorusage import planned
Devin	nowassesssessionsdoctorreader planned
Muse Code	nowassesssessionsdoctorreader planned

Frequently asked questions

$ what is decadra

A command line tool that answers one question per AI coding provider: does the plan you pay for fit the usage you actually put through it. It reads the session logs on your machine and reports each provider on its own.

$ does it send anything over the network

No. Every read is local, and ccusage, which Decadra bundles for token and cost math, always runs with --offline. There is no opt-in price refresh yet either. Decadra makes zero network calls today.

$ which providers work today

now, assess, sessions, alerts, baseline, and doctor work for Claude Code and Codex. Cursor, Devin, and Muse Code adapters exist and their doctor check confirms whether the local files are there, but reading usage from them (a Cursor CSV import, manual entry for Devin, a native Muse reader) is planned, not built yet.

$ how is this different from ccusage

Decadra uses ccusage under the hood for token and cost math, always offline. ccusage does not price your usage against every plan you could hold, add a per-session effort read, or give you a one-block readout you can run from inside an agent's shell. That plan-fit layer, and decadra now, are what Decadra adds on top.

$ why won't it tell me the cheapest provider

Token counts measure consumption, not output, and the units differ between vendors. Decadra reports each provider on its own, alphabetically, and never adds or ranks usage across them. The one number that spans providers is the bill line: the subscription prices you recorded yourself, because that is money paid, not usage.

$ what does now do inside an agent

Run it from the ! shell in Claude Code or Codex and it prints one block for the session you are in: this session so far, today, month to date against your plan, pace to month end, ceiling evidence, and the last plan-fit verdict. It exits 0 on every readout, including empty ones, because the agent reads the output too.

$ is this affiliated with anthropic, openai, or cursor

No. Decadra is an independent project that reads local log files those tools already write. It is not endorsed by or affiliated with any of them.