Session / weekly / Opus limits · token spend · cost estimates · per-session history — all read from local logs, refreshed on a timer, living quietly in your menu bar.
- What's left, at a glance. Mirrors Claude's
/usageview — three meters for Session (5h), Week · all models, and Week · Opus, each with real usage and a live reset countdown. - Real token data. Exact per-request token counts parsed straight from Claude Code's session logs — input, output, and cache read/write.
- Cost estimates per model (Opus / Sonnet / Haiku) from standard-tier pricing.
- 7-day spark chart + all-time model breakdown + recent-session history.
- Live vendor data (optional). Add a GLM Coding Plan or Anthropic admin API key — stored encrypted, machine-bound — for real GLM 5h/weekly quota and org-level Anthropic cost.
- Always current. A Rust timer re-scans and pushes fresh data to the UI — auto-refresh interval is configurable in Settings (default 30s), applied live without a restart. No frozen snapshots.
- Stays out of the way. Menubar-only (
LSUIElement), click-to-toggle dropdown, single-instance, launch-at-login. - Self-updating. Signed auto-updates via the Tauri updater — an in-app "Update & restart" banner appears when a newer build ships.
|
Overview — limits, reset timers, week chart, model split |
 Sessions — recent sessions with project, model, tokens, cost |
 Providers — connection status + live vendor usage |
 Settings — plan tier + encrypted API keys |
A thin React frontend that only renders, and a rich Rust backend that does all the work.
┌──────────────────────────────┐ ┌─────────────────────────────────────┐
│ React frontend (src/) │ │ Rust backend (src-tauri/src/) │
│ │ │ │
│ useUsage / useTauriCommand │◀──────▶│ commands/ invoke handlers │
│ Meter · WeekChart · Settings │ invoke │ scanner/ log aggregation │
│ │ + event│ vendors/ z.ai + Anthropic clients │
│ renders the snapshot │◀──────▶│ encryption/ AES-256-GCM key vault │
│ │ usage- │ settings/ · state/ · storage/ │
│ │ updated│ tray.rs menubar dropdown │
└──────────────────────────────┘ └─────────────────────────────────────┘
▲ timer (cfg) ▲ on-demand refresh
│ │
~/.claude/projects/**/*.jsonl · ~/.zai/zai-mcp-*.log
The backend scans logs (off-thread via spawn_blocking), optionally fetches live vendor data, merges one UsageSnapshot, and emits usage-updated to the UI.
| Metric | Source | Real? |
|---|---|---|
| Claude token usage (session/week/model) | ~/.claude/projects/**/*.jsonl |
✅ exact |
| Claude cost | derived from per-model pricing | ≈ estimated |
| Reset countdowns | computed from log timestamps | ✅ real |
| "% left" ceilings | editable plan tier (Pro / Max 5× / Max 20×) | ≈ estimated* |
| GLM token/cost (local) | ~/.zai/*.log — lifecycle only |
❌ shown as — |
| GLM 5h/weekly quota (with key) | z.ai monitor API (/api/monitor/usage/quota/limit) |
✅ real |
| Anthropic cost (with admin key) | Anthropic Admin Cost API | ✅ real (org-level) |
| Copilot premium-request quota | GitHub copilot_internal/user (your editor / gh token) |
✅ real (per-user) |
* The Pro/Max subscription "weekly % left" has no public API, so ceilings are estimates you set by picking your plan. The Anthropic admin key reports org-level cost, not the subscription quota.
npm install
npm run tauri dev # develop
npm run tauri build # bundle an unsigned .app + .dmg (local testing)Heads-up: if your shell sets
NODE_ENV=production, install withNODE_ENV=development npm install --include=devso the dev toolchain is included.
For a .dmg that installs cleanly on any Mac (no Gatekeeper warnings), it
must be signed with a Developer ID cert and notarized by Apple:
cp .env.example .env # fill in your signing identity + notarization creds
./scripts/release-mac.shSee docs/RELEASE.md for the full runbook (certificates, notarization credentials, verification, universal builds, troubleshooting).
- Auto-refresh interval — choose 10s / 15s / 30s / 1m / 2m / 5m in Settings (default 30s); takes effect on the next cycle.
- Plan tier — pick Pro / Max 5× / Max 20× from the header dropdown; it sets the limit ceilings and persists.
- API keys (Settings tab) — optional z.ai and Anthropic admin (
sk-ant-admin…) keys for live vendor data. - z.ai endpoint — editable; confirm it against your account's billing API.
API keys are encrypted with AES-256-GCM using an Argon2id-derived key whose password is this machine's UID — so a settings.json copied elsewhere can't be decrypted. Keys never leave Rust in plaintext and are never exposed to the frontend (which only sees …KeySet booleans).
agent-status/
├── src/ # React frontend (thin)
│ ├── hooks/ # useTauriCommand, useUsage
│ ├── components/ # Meter, WeekChart, VendorCard, Settings
│ └── styles/app.css
└── src-tauri/ # Rust backend (rich)
└── src/
├── commands/ # invoke handlers (collect = scan + vendor)
├── scanner/ # log → UsageSnapshot aggregation
├── vendors/ # z.ai + Anthropic API clients
├── encryption/ # at-rest key vault
├── settings/ state/ storage/
└── tray.rs # menubar icon + dropdown
cd src-tauri && cargo test --all # 12 tests: scanner, encryption, vendor parsersCI runs the suite on macOS / Windows / Ubuntu (.github/workflows/unit-tests.yml).
- Icon lives at
src-tauri/icons/icon.svg→icon.png; re-runnpx @tauri-apps/cli icon src-tauri/icons/icon.pngafter editing to regenerate every size. - Bundle identifier is
com.dennisrongo.agentstatus— change insrc-tauri/tauri.conf.jsonif distributing under a different org. - Signing, notarization & auto-updates for distribution — see docs/RELEASE.md.
- Live vendor endpoints are best-effort and unverified offline — confirm against your accounts on first run.