App Usage

USD per app — Sum of `session.cost` for all sessions classified to that app. Cost is OpenRouter-computed from tokens (input $0.98/M, output $3.08/M, cache $0.182/M). FIRELabs dominates because it uses NVIDIA NIM models via the optimizer.

Sparklines — Each row shows weekly sessions (bars) + tool calls (dots) for the last 12 ISO weeks. Weeks with no data are shown as empty — gaps mean you didn't work on that app that week. Hover/tap to see counts. FIRELabs + Watching = consistent weekly engagement.

First / Last — When you first and most recently touched each app (non-archived sessions). 30d Sessions / Tools — Sessions and tool calls in the last 30 days. Amber = stale (≤2 sessions & ≤100 tools in 30d). Apps with zero 30d activity are candidates for decommissioning.

What slips through — Sessions whose titles didn't match any classifier pattern. Titles like "Markdown file review" or "Greeting" are noise; specific ones like "My New App feature" indicate a missing pattern. Edit `/usr/local/bin/status-backend.pyd` `_APP_TITLE_RULES` to add new regexes — the script runs every 60s and stats.json refreshes automatically.