Repository navigation
MacFleet v0.7.0 — fleet panels and token usage
MacFleet v0.7.0 — clearer fleet health and token usage
What's new
- A sparkDash-style overview with clear CPU, GPU, memory and storage bars, focused computer details, search navigation, three themes and a compact mobile menu. The overview's CPU squiggles are removed.
- Optional 24-hour or 7-day hourly history, showing average/peak CPU, a consistent 0–100% scale and missing-hour coverage.
- Separate local-model, cloud/subscription and unclassified Hermes token records. Recorded input, output, cache-read and cache-write counters remain separate; benchmarks are not treated as live traffic.
- A read-only local token publisher and a prepared one-shot Windows pilot collector with an offline validator.
Accuracy fixes
- Storage used is total capacity minus available space. Raw APFS filesystem-used is shown separately in Details.
- Missing or malformed telemetry remains unknown/unverified, not healthy or zero. Running a backup does not imply a completed backup.
- Token timestamps now sort correctly across whole-second and fractional timestamps. The Windows validator accepts native seven-digit .NET fractional timestamps.
Version and coverage
- Viewer: v0.7.0. Health collector, probe and endpoint: v0.6.0, unchanged. This update does not require a collector, sampler or endpoint restart.
- Token records cover retained Hermes sessions on the publishing computer, not all fleet computers, Codex desktop usage, daily allocations or provider invoices. No dollar cost or combined input/cache total is inferred.
- The Windows pilot is prepared, not installed or enrolled. No Windows device is counted as healthy by this release.
- Keep dashboards on their existing private loopback/tailnet address. Live telemetry, host configuration, credentials, prompts and session identifiers are not release assets. The source archive contains code and documentation only.
Verification
- 48 Python unit tests passed on local Python 3.9 and the default Python runtime.
- 33 collector self-checks and 11 JavaScript model/release tests passed.
- Viewer syntax, staged-source secret scan, Git-history secret scan and source-archive privacy inventory passed.
- Desktop/mobile smoke checks verified token filtering, history controls, search navigation, the mobile menu and guide-link copying, with no captured console errors or warnings.
- GitHub CI passed for the release commit: viewer checks, Python 3.9/3.13 unit tests and self-checks, and secret scanning. JavaScript checks are now included alongside the existing Python matrix.
Upgrade
Update the checkout/source files and refresh open browser tabs to load the new viewer. Node.js 22 is needed only for development tests, not for running the viewer or health collector. Token publishing is an explicit local integration; the release does not install a new scheduler or enroll machines.
See the changelog and Windows pilot guide for details.