Lancetta

What makes it different

Lancetta is a native macOS menu-bar app that tells you when to push your coding agents and when to ease off. It watches the quota readings of Codex and Claude Code and turns them into plain guidance: go all in, lower the effort, hand work to a lighter model, move to the other agent, or trim what's loading. The app is free (version 0.61.0, macOS 15.0 or later, Universal binary, signed and notarized), and reading a quota costs nothing — no model is asked, no token is spent.

It works from every quota reading plus a scan of what your agents load at each start. When a 5-hour Codex window is about to reset with most of it unused, Lancetta says so — quota gone for good otherwise — and suggests starting the long task now and raising the effort if it was set low, while it keeps watching the pace. Before an agent runs dry, it warns in time: pausing sessions that can wait, lowering effort, switching routine work to a lighter model, or moving to the other agent while it has room. The weekly window gets the same treatment, because running out on a Tuesday costs days and only a series of readings sees it coming.

Every suggestion shows its working. The panel's 5-hour bar carries on faint to where the last hour's rate leaves it at reset, turning amber with a mark only when that pace would stop the agent short; a Limits view charts both windows — the week with its nights and 5-hour sessions, the five hours with a cone of where the rate leads — so you can check exactly what a suggestion rests on. Today a suggestion reasons from this window's own rate; next, it will reason from an average across your own days to say when to start and when you'll probably stop.

Beyond quota, Lancetta does maintenance on how your agents are set up. Every session pays tokens for its instruction files — CLAUDE.md, AGENTS.md, global rules, memory index, one line per skill — so it scans each repository Claude Code or Codex has worked in recently plus the folders above, measures what loads at each start against Claude Code's own warning thresholds, and reports 22 checks each with its reason. Fixes that need no choosing are shown in full before being applied, one at a time or all at once; the rest open in Claude Code or Codex with the prompt already written.

It also keeps your sessions and your disk tidy. All your Claude Code and Codex sessions become post-its on a wall — the ones whose turn is yours first, then the ones still working — and a background session can open in your terminal from there. And since agents never close what they leave behind, Lancetta finds the orphaned background trees in every folder an agent worked in (the site cites one Mac with 28 trees holding 2.68 GB), shows the list, and suggests reclaiming the memory before it closes anything.

On a MacBook Pro Lancetta also lives under the notch: one bar per agent, exactly as wide as the notch, turning red when its agent has stopped or will within the hour, opening the same suggestion as the panel on click. Command-O opens the full window — suggestions first, then daily tokens for both agents over 7, 30 or 90 days, every window in detail, the week that just ended, leftover processes, maintenance, and the session wall. It is built by Giovambattista Fazioli (gfazioli), a long-time open-source maker whose studio page lists Lancetta among his projects, and it is completely free with docs, FAQ and release notes on the site.