LunoPeak

Another update on what's shipped. Last time (1.4) the headline was Ollama and Cursor support. Since then LunoPeak has gone from "a dashboard you open" to "a thing that runs quietly in your menu bar and taps you on the shoulder when it matters." The big one: LunoPeak is now a proper menu-bar app and a proper window app at the same time. When the window's open you get a dock icon and working menus like any native Mac app. Close it and the app tucks into the menu bar — the dock icon disappears, but it keeps refreshing your usage in the background. Launch-at-login starts it silently in the menu bar with no window pop and no stolen focus. Click that menu-bar icon and you get a redesigned popover: a frosted, themed panel with per-provider usage bars, an insight strip that surfaces your single most-pressing window (color-coded green/amber/rose with the reset time), today + 30-day spend, and a 7-day spend sparkline. Tabs per provider, quick actions back into the app. And it'll warn you now. There's a new Alerts tab in Settings — get a native desktop notification the moment any provider's 5-hour or weekly window crosses your threshold (default 80%, escalating to a red alert at 90%+). It's deduped per provider × window × billing cycle, so you get one heads-up per window, not one every poll. Claude Code, Codex, and Cursor are all covered. Costs also grew up. Click any repo to open a full cost drill-down — breakdowns by day, by 5-hour session, by week, by weekly session, each with sessions / messages / tokens / cost. Totals now go beyond dollars (total messages, total tokens, total sessions), and every bar chart in the app picked up rich hover tooltips with real calendar dates — "Tue, May 5 · 12 sessions" instead of decoding "5d ago." A theming overhaul landed too: dark mode rebuilt f

Quick update on what's shipped since launch. LunoPeak started as a local-first dashboard for Claude Code and Codex. Three releases in, it's a lot more than that. The big one: Ollama is now a first-class assistant provider. The chat panel can run entirely on local models — no API key, no account, no cloud round-trip. Tool calls work end-to-end on Llama 3.2, Qwen 2.5 Coder, and Mistral, so the assistant can still pull cost breakdowns, list repos, run hygiene checks, and inspect sessions even when it's running 100% offline. If your Ollama daemon is up when you first launch the app, it gets auto-adopted as the default before you ever touch Settings. Cursor joined the agent lineup too. Claude Code, Codex, and Cursor all show up in Live, Setup, Repos, and Settings — one window, every agent, bucketed by repo so they don't bleed into each other. A few other things worth calling out: - A combined provider + model picker in the chat header — flip from Claude Sonnet to a local Qwen and back without leaving the conversation. - Drag-and-drop dashboard. Rearrange the cards, layout persists per device. - GitHub PRs in the Work Graph view, per-file dirty state on the Repos view, and a "spend this month" number that's actually month-to-date. - Dashboards refresh themselves now — 60-second cadence on the time-sensitive queries, so the numbers tick forward instead of freezing on whatever was true at launch. Still local-first. Still no account. Still no telemetry. Still free. macOS, Windows, and Linux builds: https://github.com/idevtim/lunopeak/releases/latest Curious what other indie devs would want to see in a tool like this — what's the metric or view that's missing from your own AI coding setup?