Lantern 0.17.8

Fugu plan limits stay visible after short review commands end. Plan Usage shows session and weekly percentages left, plus the weekly reset. The reader uses the configured local profile and key. It starts no model turns. Keys never enter the shared cache.

Claude model preflight accepts valid token logins that return costs without quota buckets. It reports quota as unknown. Failed commands, malformed output, and exhausted quotas still fail.

Headless Codex jobs require an explicit folder. The shipped plugin no longer supplies personal folder, model, or tab defaults. Private user settings remain available.

The background version check replaces its file in one step. Readers keep the previous complete line until the new line is ready.

Docs and runtime instructions are updated. Independent reviews, local gates, and Linux, macOS, and Windows CI passed on the reviewed heads. The final merge has the same file tree as the reviewed Fugu PR.

Changes: #53, #47, #59.

Lantern 0.17.5

Plan Usage no longer warns about unreadable extra tabs when the service has measured limits. It still shows stale readings, failed updates, and required sign-ins.

Claude uses the total weekly reset when model limits have the same percentage within one login. A lower model allowance still sets the percentage and reset. Separate logins keep their own reset times.

Percentage columns align to the right. Weekly reset times use normal text contrast. Docs and runtime instructions are updated.

Both independent reviews passed. All local checks and Linux, macOS, and Windows CI passed on the reviewed head. The merge has the same file tree.

Changes: #58

Lantern 0.17.4

Plan Usage now shows weekly reset times beside the percentages. Times use the local time zone. The reset belongs to the limit that sets the displayed weekly percentage.

Each service keeps one row. Session and weekly allowances stay as percentages. Tied weekly limits with different reset times show Varies. Missing or expired reset times show --. Panes under 39 columns keep the percentages and ask for more width.

The Field Status and usage layout stays the same. The guides, runtime instructions, and changelog describe the new column. Provider reads are unchanged.

Lantern 0.17.3

Field Status now opens in the top two thirds of the right column. Plan Usage opens in the bottom third. Chat stays on the left. Setup repairs the previous order without changing focus or restarting verified watchers.

Plan Usage shows one row per service. The table has three columns: Service, Session left, and Week left. Values are percentages. IDs, model names, plan codes, credits, bars, and reset times stay hidden. Missing values show --. Stale readings have a warning. Each period shows the lowest remaining percentage across the service's open login profiles and quota pools.

Google reads the native Antigravity usage command with zero model turns. Grok reads the saved OAuth plan when an inherited API key is present. Fugu reads Sakana's usage API with the key selected by the open Codex profile. Both modern profile files and legacy profiles work. No quota read sends a model prompt. Keys never enter the shared quota cache.

The guides, runtime instructions, and changelog now describe these changes.

Lantern 0.17.2

Plan Usage now shows service, session percent left, weekly percent left, and reset times in ET. The table hides account and model IDs, plan codes, credits, source details, update age, and allowance bars. Narrow panes put reset times below percentages.

Each period shows its lowest remaining allowance and the reset for that allowance. Separate logins keep separate values. Missing values show --. Stale values and past reset times keep warnings. Low allowances keep warning colors. Plain output and NO_COLOR keep the same text.

The Google and Grok source restrictions remain. This release changes the display.

The docs, runtime instructions, changelog, and plugin version use 0.17.2.

Validation: two independent reviews approved the final commit. Local tests and GitHub checks passed on Linux, macOS, and Windows. Display checks covered widths from 20 to 88 columns.

Changes: PR #55.

Lantern 0.17.1

Lantern now opens chat and both readouts in one tab. Chat is on the left. Plan Usage is on the top right. Field Status is on the bottom right. Existing verified readout watchers move into this layout without a restart. Setup keeps focus and preserves existing pane sizes.

Plan Usage shows one table row per service with an open agent tab in any workspace. Idle and Done tabs count. The table shows allowance left, reset times, compact allowance bars, and warning colors. Separate accounts keep separate values within the service row. Unknown accounts and stale values stay visible. The display omits agent details and tab counts.

Both readouts fit their pane. Use Up/Down or Page Up/Page Down to scroll. Press q to exit. Terminal settings restore on exit. Narrow panes, plain snapshots, and NO_COLOR are supported.

The usage collector reads OpenAI, Claude, and Grok limits from verified sources. Claude and Grok need the pinned CodexBar CLI on macOS or Linux. Antigravity shows Account unknown. Grok also shows Account unknown when its process has a credential or base URL override. Windows shows Account unknown for OpenAI and Dependency unavailable for Claude and Grok.

The README, guides, runtime instructions, changelog, and plugin version now use 0.17.1.

Validation: two independent reviews approved the final commit. Local tests and GitHub checks passed on Linux, macOS, and Windows. Live Herdr checks covered startup, readout reuse, restoration, scrolling, watcher identity, and focus.

Changes: PR #52 and PR #54.

Lantern 0.16.1

Changed

• Green is two answers on the exact head: the PR can merge (required checks
  passed or were skipped, MERGEABLE, not BLOCKED or BEHIND), and it
  was tested (a Local tests passed on <head SHA> record from a repo
  member listing every gate as passed, passed full-suite jobs on a
  Dependabot PR, or passed release checks).
  Socket, Vercel, and skipped checks never prove that tests ran.
• Post-merge steps follow the landing path. A merge into main keeps the
  tag/release, deploy check, and main pull. A merge into dev checks the
  staging deploy and pulls dev. A release or hotfix into main gets a
  same-day back-merge PR into dev. Publishing the GitHub version after an
  everyday merge stays part of the kickoff's merge authority where main
  is the default branch; a release into main while dev is the default
  needs explicit release authority. The close bar reads the merged branch
  from the PR base.
• Session cleanup blocks on a missing required check, not on a repository
  with no checks. gh pr checks reporting no checks is not a failure.
• Bot reviews that ran inside PR CI (Gemini auto-comments, Claude review
  workflows) are no longer expected once a repo stops PR tests.

Lantern 0.15.0

Fugu seats use the installed Codex Fugu profile.

• The default is regular fugu at high effort.
• Ultra prefers fugu-ultra-v2.0, then fugu-ultra, then fugu-ultra-v1.1.
• Effort max stays on the first of those rows that lists it.
• The Fugu Max model fugu-max is selected only when named.
• A missing codex-fugu launcher stops the seat. Lantern does not start plain Codex.

Reopen Lantern after you update so the new seat rules load.

Usage guide

Lantern 0.14.1

Claude folder trust no longer confirms No, exit.

• When the card highlights No, exit, the gate sends Down.
• Enter follows only when the marker is on Yes, I trust this folder.
• The older card, with trust already selected, still gets one Enter.

This includes the folder-trust fix from Aummadour.

Reopen Lantern after you update so the new gate loads.

Usage guide

Lantern 0.14.0

Model routes follow the live catalogs.

• Claude Opus pins the resolved id. On this catalog that is claude-opus-5-5[1m].
• Bare Grok seats use Grok Build. The default is grok-4.7-build-fast at medium effort.
• "Cursor" or "in Cursor with Grok" still selects the Cursor CLI.
• Codex accepts gpt-6-sol and gpt-6-luna. Bare sol, bare luna, and bare gpt-6 stay ambiguous.
• Cursor accepts Codex 5.3 and Opus 5.5 ids.

Reopen Lantern after you update so the new seat rules load.

Usage guide