Track your Claude Code spend, broken down by git branch, in one command. No account, no API key, nothing in your request path.
npx goei-sync reads the session logs Claude Code already writes under ~/.claude/projects, prices every response at published list rates, and prints how much each project and git branch has cost you. It is the per-branch view ccusage and the native /cost command do not give you, and it runs entirely on your machine.
npx goei-syncClaude Code local usage by git branch (2026-07-10 to 2026-07-23)
Local usage value at list prices. Not your provider bill.
Top branch: web/main $347.70 (27% of $1265.77 across 54 branches)
PROJECT BRANCH TOKENS USD
web main 458,244,888 $347.70
web fix/login 45,402,797 $26.52
api main 33,271,226 $23.78
TOTAL 536,918,911 $398.00
By model:
opus-4-8 500,000,000 $360.00
sonnet-5 36,918,911 $38.00
These are local usage values at published list prices, not a provider bill: a real invoice reflects your subscription, negotiated rates, and credits this cannot see. Add --json for machine-readable output, or --days 7 / --since 2026-07-01 to scope the window.
For the full walkthrough, see the tutorial How to track your Claude Code spend over time, per project and branch.
npx goei-sync wrappedwrapped turns the same local logs into a single card you can screenshot. The headline is the number only local re-pricing can compute: the list-price value of the Claude Code usage that ran through your machine, plus what prompt caching saved you against paying for those same tokens as fresh input.
┌──────────────────────────────────────────────────────────────┐
│ CLAUDE CODE WRAPPED 2026-06-28 to 2026-07-19 │
│ Local usage value at list prices. Not your provider bill. │
├──────────────────────────────────────────────────────────────┤
│ │
│ $190.45 of Claude Code, at list prices │
│ across 4 active days, 3 projects, 5 branches │
│ │
│ Prompt caching saved $682.05 │
│ Biggest project web-app $155.00 81% │
│ Biggest branch web-app/main $110.00 │
│ Priciest day 2026-06-28 $110.00 │
│ │
│ Model mix opus-4-8 89% sonnet-5 10% haiku-4-5 1% │
│ │
├──────────────────────────────────────────────────────────────┤
│ See your own: npx goei-sync wrapped goei.roninforge.org │
└──────────────────────────────────────────────────────────────┘
The "prompt caching saved" line is the honest net figure: what those cached tokens would have cost billed as plain input with no cache, minus what they actually cost. It is a value comparison at list prices, not money credited to a bill.
One machine is one machine. To combine every developer and every machine into one deduplicated rollup, and to re-price 12 months of history at the rates that were live in each past month, sync to Goei:
npx goei-sync --token goei_dt_your_device_tokenSign in at goei.roninforge.org with a magic link, create a device token under Settings then Device Tokens, and pass it with --token (or set GOEI_DEVICE_TOKEN and run npx goei-sync sync). Goei prices the tokens on its side and dedupes across your machines. It is free for a single developer. See a live dashboard with no signup at goei.roninforge.org/demo.
npx goei-syncandnpx goei-sync wrappedread the raw~/.claude/projects/*.jsonllogs directly, so they attribute every message to its git branch and price the exact 5-minute vs 1-hour cache split. They price locally with the open ai-price-index dataset, dated so each response is valued at the rate that was live the day it ran.npx goei-sync --token ...(sync) reads your usage through ccusage and sends daily, per-model token counts to Goei, which prices them. ccusage reports a single combined cache-creation figure, so synced dollars value cache writes at the 5-minute rate; the local report and card above are exact.
Either way, goei-sync reads only what Claude Code already wrote to disk. It never sees prompts, responses, or API keys.
report (default) local spend by git branch; no token needed
wrapped one shareable card summarising your usage; no token
sync push daily rollups to Goei; uses --token or GOEI_DEVICE_TOKEN
--token <t> Goei device token; routes to sync (env: GOEI_DEVICE_TOKEN)
--days <n> limit to the last n days
--since <date> limit to YYYY-MM-DD forward (overrides --days)
--json print report or wrapped as JSON instead of a table (local only)
--machine <id> machine label for dedupe on sync (default: this host's name)
--endpoint <url> ingest endpoint (default: https://goei.roninforge.org/api/ingest)
--show-payload print the exact JSON that sync would send, then exit
--dry-run run the sync summary but send nothing
-h, --help show this help
report and wrapped are local-only and refuse any sync flag, so they can never send anything. Run npx goei-sync sync --show-payload to read the exact request body before you ever send a token. Every field it contains is a token count, a date, or a model name. There are no prompts and no keys in it.
Both read the same local Claude Code logs with no API key. ccusage gives you daily, monthly, and per-session totals. goei-sync adds the two things it does not: a per-git-branch breakdown of where your spend actually went, and a one-command push to a shared team dashboard. Use goei-sync when you want to know which branch or project burned the budget, or when more than one developer needs to see the total. A side-by-side breakdown lives at Goei vs ccusage.
Both feed the same Goei dashboard from the same local logs, with no API key. On a machine that syncs, run one of the two, not both: a day synced by both would be counted twice. budgetclaw adds always-on background tracking with per-project and per-git-branch attribution, plus hard spend caps that can stop a runaway agent before the bill lands. Use goei-sync for a quick local look, a shareable card, and a dependency-free push; use budgetclaw when you want continuous tracking and enforcement. If you want to stop a runaway agent, not just measure it, start with the tutorial How to set a hard spend cap on Claude Code.
goei-sync is a reporter, not a proxy. It does not sit between your editor and any AI provider, and it touches no API traffic, prompts, responses, or keys. It reads the usage Claude Code already wrote to disk.
Node 18 or newer.
MIT. Part of RoninForge, which builds honest, local-first cost tools for developers working with AI coding assistants. More on tracking and capping Claude Code spend: Goei, budgetclaw, and the Claude Code cost tutorials.