One subscription, many machines. UsageFleet reports Anthropic's own 5-hour and weekly utilization and splits it across the device groups you define, so you can see which laptop, desktop or server burned the window.
usagefleet.com — sign in with GitHub or Google. Free for one device.
The percentages are Anthropic's, not an estimate: a small collector on each
machine reads them from Anthropic's own usage endpoint (the one Claude's
/usage screen uses) with the Claude login you already have there, falling
back to the anthropic-ratelimit-unified-* response headers for API keys.
UsageFleet only splits them — each rise of the percentage is attributed to
the groups active when it happened, weighted by estimated cost; usage no
monitored device produced shows up as "Unattributed" instead of being
redistributed.
A group's number is measured against its own slice of the account (1/N with N groups), so "62%" means that group used 62% of its budget — not 62% of yours.
- Sign in at usagefleet.com.
- Devices → add a device, pick a group, copy the token (shown once).
- On that machine:
npm i -g @usagefleet/cli
usagefleet login uf_xxxThe dashboard fills in within a minute. Same two commands on macOS, Linux and Windows; the collector then runs in the background at login and updates itself. Full manual: apps/cli/README.md.
Phones are out of scope — the Claude mobile app keeps no local usage logs.
Read-only tailing of local agent logs: Claude Code
(~/.claude/projects/**/*.jsonl), Claude Desktop agent-mode sessions, and the
pi agent (Anthropic-provider records
only).
Uploaded: token counts, model, session id, hostname, working directory, git branch. Never uploaded: prompts, responses, file contents, or your Claude credentials — the limits reading is signed locally and only the resulting percentages leave the machine.
| Plan | Devices (= max groups) |
|---|---|
| Free | 1 |
| Solo | 2 |
| Fleet | 8 |
| Custom | 10–200, priced per device |
Prices and checkout live on usagefleet.com. Cancel any time; the free plan keeps working.
- Alerts — desktop notification the first time a window crosses 80% / 95%.
- Guard —
usagefleet guardis a Claude Code hook that refuses new prompts once a group is over a window it blocks on. Fails open on any problem. - Spend — cost estimate per window, group and model, from published pricing.
- History — past 5-hour blocks and weeks.
Pick a period, a metric (billable, total, input, output, cache-read or cost) and what to split the bars by — group, model, device or source. + Filter narrows by any of the same four, so "Opus on the two work desktops, last 90 days" is a few clicks; with nothing filtered the controls stay out of the way. The ledger under the chart totals the same slice.
Two Bun workspaces, one lockfile: apps/web (TanStack Start + better-auth +
Drizzle + Postgres) and apps/cli (@usagefleet/cli, zero runtime deps).
docker run -d --name usagefleet-db -e POSTGRES_DB=app -e POSTGRES_USER=app \
-e POSTGRES_PASSWORD=app -p 5432:5432 postgres:17-alpine
cp .env.example .env # one .env at the repo root serves compose and dev
bun install
bun run db:migrate
bun run dev # http://localhost:3000
bun run test # usage math + collector unit testsSchema changes: edit apps/web/src/db/schema.ts, run bun run db:generate,
commit the SQL under apps/web/drizzle/.
| Path | What |
|---|---|
apps/web/src/db/ |
Drizzle schema (auth, subscription, groups, devices, usage) |
apps/web/src/lib/usage/ |
fold, windows, pricing (pure, tested); the group split lives in lib/data.ts |
apps/web/src/routes/api/v1/usage.ts |
ingestion endpoint (x-api-key, dedup on uuid) |
apps/web/src/routes/_dash/ |
dashboard, groups, devices, billing, settings, admin |
apps/cli/ |
the usagefleet collector |
docs/architecture.md— request flow, data model, auth, billing.docs/usage-math.md— fold, windows, the group split, pricing.docs/collector.md— CLI internals.CONTRIBUTING.md— setup, style, what to read first.SECURITY.md— reporting a vulnerability.
GPL-3.0-or-later. Run it, modify it — published modifications stay under the same license.