MacFleet Dash v0.10.0 — SparkDash insights

MacFleet Dash v0.10.0

Optional read-only SparkDash insights, plus the previously prepared 0.9.5
coverage and saved-profile improvements.

New

• AI engine cards: reported model/backend, decode and prefill rate, running
  and queued requests, occupied/total slots, and explicit unavailable values.
• Engine token windows: 1h, 6h and 24h, grouped by machine and model. These
  are hourly UTC buckets with a partial current hour, not exact trailing-minute
  windows. Prompt includes cached tokens; cache is not added a second time.
• Resource and AI history: 1h, 6h, 24h, 7d and 30d, fixed-scale CPU/GPU
  averages/peaks, missing-data gaps and inspectable headroom/network/AI rates.
• Coverage & repairs and per-computer/saved-profile Hermes counter selection.
  Copied or migrated computer histories are never combined into a fleet total.

Reliability and privacy

• Read only existing SparkDash cached metrics, token ledgers and history.
  No extra engine polling, prompts, benchmarks, model/settings changes,
  notifications, pricing downloads or new background services.
• Reuse SparkDash's reconnect-aware rates and durable ledger. Failed reads
  preserve the last successful receipt and show retained values as unverified.
• Bound browser snapshot reads to avoid connection resets when optional
  sources arrive together. Preserve partial-profile usage coverage.
• Keep engine-side traffic separate from Hermes/cloud client counters and
  billing. Cached-metrics receipt time is not an individual probe timestamp.
• Explicit address-checked links do not add duplicate computers. Worker
  hardware is not treated as a second token-producing engine.
• Public downloads contain source and examples only: no operator inventory,
  private source URLs, mappings, credentials or runtime readings.

Installation

Refresh the viewer after updating source. SparkDash remains optional; configure
your own existing private source using the README. Use your existing snapshot
refresh for the optional publisher. History must already be enabled in SparkDash.
No Tailscale enrollment or network-policy changes are included.

Collector source v0.6.2 is included. An already-running health collector retains
its actual loaded version until its normal service reload; viewer and optional
publisher updates do not restart AI services or the private telemetry receiver.

Verification

116 Python tests, 67 viewer tests and 33 collector checks passed locally.
Desktop/mobile browser checks exercised machine filters, all three token
windows, all five history windows, failure recovery, existing usage controls
and a clean extracted installation. The actual 60-file source archive passed
privacy checks and secret scanning.

Rollback

Restore v0.9.4 source while preserving private configuration and runtime state,
then refresh the browser. Disable the optional connector with enabled = false
in your own SparkDash configuration section if needed. Reload only your existing
MacFleet collector through your normal service manager when adopting collector
source changes; no AI-service restart is required.