Metrics B2BSolutionsMetrics PersonalCLIDocumentation

runs locally · no account · no network

How much did this ticket cost?PanDev has the answer.

Your provider bills a single figure a month. PanDev breaks it down by task, branch, model and file — right on your machine, and not a byte leaves it.

$npx pandev@beta

Node 20+ · macOS, Linux, Windows · nothing else

works withClaude CodeClaude CodeCodex CLICodex CLIOpenCodeOpenCodeZCodeZCode

where the tokens go

Without PanDev — one number. With PanDev — the breakdown.

without PanDev

$2,140

all your provider's console shows

········
········
········

with PanDev

$2,140

the same money — by task and by prompt

PAY-1204
$842
10:41 · add retry with backoff, keep it idempotent$186
09:03 · write the migration and a rollback for it$91
14:20 · cover the duplicate-delivery case with a test$68
+ 8 prompts$497
AUTH-88
$488
INFRA-31
$301
PAY-1198
$203

one command

One run. Zero setup.

The logs are already on your disk. PanDev reads them, prices them at model rates and puts a ticket number on every line — offline, nothing leaves your machine.

90%
of spend lands on a ticket — branch → issue key
0.5s
for a full report on 100 MB of logs — nothing is uploaded first
4
ways to slice it: task · file · model · surface
90d
of branch history recovered from git reflog, not guesswork

the part nobody else has

Open a ticket. See every prompt it took.

Not just a total — the actual sequence. What you asked, what each ask cost, how many rounds it took, which files it moved.

Time that means something.

Active time excludes the hours you were away — and you still see wall clock next to it.

Cost per prompt.

One vague ask can be half the ticket. You find that out the same day, not at month end.

Model and surface.

Which LLM did this ticket — and whether it ran in VS Code, JetBrains or a terminal.

per-file breakdown

The file that took 27 rounds to get right.

What each file cost, how many rounds of edits it took and the price of the prompts behind it. The link is exact: every tool call sits in the same log record as its cost.

PanDev prints what share of spend this view covers — a slice is never passed off as the whole.

pandev web

And a beautiful dashboard. One command.

pandev web builds a self-contained page and opens it in your browser. Offline, no server, no uploads — it works on a plane.

privacy

It can't leak what it never sends.

There is no network code in PanDev — not disabled, absent. And you can verify that, not take it on faith.

Verify it, don’t trust it.

Run tcpdump next to it — silence. Builds are reproducible, checksums are published, pandev privacy prints the binary’s sha256.

No account, ever.

Nothing to sign up for, nothing to revoke. Delete the binary — no trace left.

Your prompts stay yours.

Read only when you open a ticket, printed to your terminal — and never sent anywhere.

honestly

Where PanDev fits

Counters already exist and they're good at counting. This is a different question.

Question you're askingusage counterstime trackersPanDev
How much did I spend yesterday?yesnoyes
How much did this ticket cost?nonoyes
Which prompt burned the budget?nonoyes
How long was I in the editor?noyesno
Which file took the most rounds?nonoyes
Works with no accountyesnoyes

questions

Straight answers

Is this a productivity tracker?+

No. It doesn't score you, rank you, or report to anyone. There is nobody to report to — it runs on your machine and talks to nothing. It answers one question: where the money went.

How does it know which ticket a spend belongs to?+

From the branch name: feature/PAY-1204 → PAY-1204. And it uses the branch you were on at the time of the work, recovered from git reflog — not whatever branch you happen to be on now. That distinction is the difference between a real number and a plausible one.

What if I work outside a repo, or on a branch with no ticket?+

Then it says so. Unattributed spend is split into “in a repo, branch has no key” (rename the branch and it attributes itself) and “outside any repo” (nothing to attribute to, ever). You always see what share is which.

Which agents does it read?+

Claude Code and Codex CLI today, from the logs they already write. It also records which surface the agent ran in — terminal, VS Code, JetBrains, desktop — so you can compare them.

Do I need to be on an API key? I'm on a subscription.+

Subscriptions are handled properly. Instead of showing dollars you never spent, PanDev shows payback: what the same tokens would have cost at API rates versus what your plan cost for the period. That's detectable only on your machine — the server can't tell the two apart.

Does it upload anything? Anything at all?+

No. There is no HTTP client in it, no URL, no telemetry, no update check. Verify the behavior instead of the promise: run tcpdump next to it, and check the sha256 that pandev privacy prints against the published checksums.

and then

Works for one. What about the ten of you?

PanDev shows your machine. PanDev Metrics answers the same question for the whole team — every repo, every agent, every editor, the same task-first attribution — plus the delivery metrics around it.

Explore PanDev Metrics for teams

You already have the data.It takes one command to read it.

$npx pandev@beta

Node 20+ · macOS, Linux, Windows · nothing else

Find it useful? Star it on GitHub

GitHub8

Screenshots on this page use invented data.