Port upstream 0.66.0: OpenAI daily usage chart UI (stacked on #699)

[Finesssee]

Summary

Part (b) of the deferred upstream CodexBar 0.66.0 "OpenAI per-day usage chart": the UI. Part (a) (Rust model, provider fetch, bridge DTO, types/bridge.ts) is #699; this PR is stacked on it and targets port/micro-0.66.0-openai-daily-usage-core.

• New OpenAIApiUsageChart: Cost | Tokens toggle, one bar per UTC day, a day detail panel (cost, requests, input / cached input / output tokens, line items, models) and a footer with the window label and total.
• Tray / pop-out MenuCard: collapsed "Daily API usage" section for openaiapi only, hidden on error, in compact overview, and when daily is empty.
• Settings > Providers: ChartsSection draws the same chart for openaiapi from the new ProviderDetail.openAiApiUsage field.
• BarChart gains an optional controlled selection prop (listbox/option semantics, roving tabindex, arrow/Home/End keys). Without it, behavior is unchanged.
• 13 new locale keys (OpenAIChart*) in all 8 locales, LocaleKey, and i18n/keys.ts.

Approved design (summarized)

From design-openai-per-day-chart.md, implemented as written:

• Reuse the existing dependency-free SVG BarChart; no new charting or other dependency.
• Cost | Tokens toggle as a tablist with roving tabindex; bars are a roving-tabindex listbox; hover or focus selects a day.
• Detail panel under the chart is an aria-live=polite region; line items and models are capped at 5 rows plus "+N more".
• Footer shows the window label ("Today" / "Last N days") and the total for the selected metric.
• USD only, matching the Admin API; provider id is openaiapi (the id openai aliases Codex).
• No new Settings tab; openaiapi stays out of PROVIDER_CHART_DATA_IDS (the local-log tabs do not apply).
• Provider data stays siloed: nothing renders for any other provider even if it carries the same payload.

Upstream reference

Pinned to steipete/CodexBar tag v0.66.0 (read only): openai.js, OpenAIAPIUsageSnapshot.swift, CostHistoryChartMenuView, docs/openai.md.

Deferred / not done

• CODEXBAR_SEED_USAGE_JSON OpenAI-API seed variant for proof harness runs.
• Currency conversion (USD only).
• OPENAI_HISTORY_DAYS setting.
• Tray chart animations are disabled (animations={false}); Settings honors enableAnimations.

Interpretation notes

• The shared 280px chart geometry cannot fit 366 bars, so the chart draws the latest 60 days; when trimmed, the footer label uses the drawn day count.
• A single-day window renders; an empty daily renders nothing in the card/Settings (the standalone component shows DetailChartEmpty).
• Added open_ai_api_usage to the Tauri ProviderDetail DTO (and TS ProviderDetail) so Settings has a data path; it is only filled when the snapshot has no error.
• The project id is not shown.
• Fluent placeholders use the existing { "{}" } convention.
• Selection is stored by bucket start so it survives a refresh.

Validation

Commands ran through the E-core wrappers with cargo +1.98.0.

Command	Result
cargo fmt --all	clean
cargo clippy --manifest-path rust/Cargo.toml --all-targets -- -D warnings	pass
cargo clippy --manifest-path apps/desktop-tauri/src-tauri/Cargo.toml --all-targets -- -D warnings	pass
cargo test -p codexbar (includes locale completeness test)	2194 passed, 0 failed, 1 ignored
cargo test -p codexbar-desktop-tauri	466 passed, 1 failed (see below)
pnpm test (vitest, full)	69 files, 429 tests passed
vitest run new/touched: OpenAIApiUsageChart, BarChart, MenuCardOpenAIUsage, ChartsSection	33 passed
pnpm run lint	0 errors; only pre-existing warnings
pnpm run build (check-locale, tsc --noEmit, vite build)	pass; 892 locale keys match between Rust and TS

Known failure, not caused by this PR: commands::tests::bootstrap_payload_exposes_every_provider_variant fails (catalog 79 vs 78 active providers). It fails identically on the unmodified base branch (804ff96a), and it reads the machine's real settings via dirs::config_dir(), so it is environment dependent.

New tests: OpenAIApiUsageChart.test.tsx (13), MenuCardOpenAIUsage.test.tsx (6, including provider siloing, error, compact overview, empty window, describeCard presence), BarChart selection mode (5), ChartsSection openaiapi (3), and a Rust test for ProviderDetail.open_ai_api_usage (default None, omitted from JSON, serialized as openAiApiUsage).

Affected areas

Tray/pop-out MenuCard, Settings > Providers detail pane, shared BarChart, locale files (8), Tauri get_provider_detail DTO, styles.css.

UI proof

Pending: coordinator will capture CUA proof on a fresh build. No desktop app was run for this PR.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Finesssee]

Thermo-nuclear review

Reviewed by Codex gpt-6-luna (xhigh); verified and validated by Claude

The PR was checked for completeness against the approved design note (chart, toggle, day detail, footer, tray/pop-out card, Settings, 13 locale keys in all 8 locales, tests). Nothing was missing. Findings:

• P1 MenuCard.tsx, IdentitySection.tsx: the OpenAI Admin project id showed in the card plan badge and in Settings identity rows even with "Hide personal info" on. Fix: mask it on both surfaces (lib/openAiApiIdentity.ts), with tests.
• P2 OpenAIApiUsageChart.tsx: the chart silently dropped history beyond 60 days, so bars and the footer total omitted data. Fix: draw the full window in a horizontally scrollable chart (BarChart scrollable prop, width-aware chartGeometry).
• P2 useChartAnimation.ts: entrance delay grew with every bar, about 8 s for a 366-day chart. Fix: only the first 30 bars stagger; test added.

[Finesssee]

Thermo-nuclear fixes landed at 813458d ("Address thermo review").

Fixed: all 3 findings above (project id masking, full-window scrollable chart, capped animation stagger).
Left: none. Note: Codex also reflowed a few unrelated lines in OpenAIApiUsageChart.tsx (formatting only).

Commands run (E-core wrappers, pnpm): vitest run full (72 files, 437 tests passed), pnpm run lint (0 errors, pre-existing warnings only), pnpm run build (pass). No Rust changed, so cargo fmt/clippy/test were not re-run. UI proof (CUA) still pending on a fresh build; the scrollable chart and masked project id are user-visible.

[Finesssee]

CUA proof

Build commit: 813458d0f9a99b359051794fd3fbf0ad6eb7275b (PR head) plus throwaway proof-only patches, never committed (reverted with git restore; diff kept outside the repo):

• Cargo.toml: [patch.crates-io] dirs = <local shim> so all config/data dirs live under an isolated CODEXBAR_PROOF_HOME (no real profile data).
• rust/src/providers/openaiapi/mod.rs: Endpoints::production() honors CODEXBAR_PROOF_OPENAI_BASE (plain-http localhost mock; no TLS or URL validation weakened) and the keyring target is renamed so the real Credential Manager entry is never read.

Data path: the real Admin fetch path (fetch_admin_usage -> history::daily_usage -> bridge -> React) against a local mock serving synthetic per-day costs and completions buckets for 30 UTC days (3 models, 2 line items, a few idle days). Dummy key only. Only openaiapi enabled, theme auto.

Commands: pnpm --dir apps/desktop-tauri install --frozen-lockfile, pnpm --dir apps/desktop-tauri run tauri:build:debug (CARGO_TARGET_DIR proof-1); launched with CODEXBAR_PROOF_MODE=trayPanel and settings:providers; driven with cua-driver call (background UIA invoke/select/set_value only; window on the second monitor, no foreground input).

#	Assertion	Result
0	No real email/account from the machine visible	PASS
1	Tray: only the OpenAI API card; collapsible "Daily API usage" section present	PASS
2	Tray section expands to Cost/Tokens tabs, 30 daily bars (2026-09-01 .. 2026-09-30), latest day selected: cost $2.92, 372 requests, input 558,000 / cached 186,000 / output 93,000	PASS
3	Tokens tab switches the axis and total: 1.6M first day, 24.2M total (matches 24,150,000 tokens on the card); Cost total $108.77 = $92.38 + $16.39	PASS
4	Settings > Providers > OpenAI API: chart section shows the same data plus "Cost by line item" (Text tokens $2.48, Cached input $0.44) and Models (gpt-5.2 420K / 240 requests, gpt-5-mini 147K / 84, o4-mini 84K / 48)	PASS
5	Mock log: GET /v1/organization/costs and /v1/organization/usage/completions, both bearer-present, 30-day window	PASS
6	Theme auto stays dark in tray and settings	PASS
7	Selecting another day by hover/focus	NOT EXERCISED (background click does not trigger hover/focus; foreground input is off limits). Covered by the PR's unit tests.

Not verified visually: the tray card's line item / model lists below the fold (background scroll is unavailable on this surface); they are shown in the Settings screenshots.

Screenshots (local, not committed), C:\Users\FSOS\AppData\Local\Win-CodexBar\port-audit\proof\701\shots\:
01-tray.png, 03-tray-daily-usage.png, 04-tray-tokens-tab.png, 09-settings-tall.png, 10-settings-tokens.png.

[Finesssee]

Validated 2026-10-02 at 813458d (PR head; base #699 at the new head f328c66 pending merge-in — see stack note):

• Reviewed against the deferred-feature scope (DONE.md item 2, OpenAI per-day chart; 0.66.0.md row 28a item 5 + the Windows design note in the PR body): the chart UI (OpenAIApiUsageChart, MenuCardOpenAIUsage, ChartsSection/IdentitySection wiring), the Cost/Tokens tabs with per-day bars, the latest-day selection, and locale keys across all 8 .ftl files match the design note; the thermo fixes at this head are included.
• cargo fmt --all --check — clean.
• cargo test -p codexbar — 2194 passed / 0 failed / 1 ignored (with the pinned 1.98.0 toolchain).
• cargo test -p codexbar-desktop-tauri — 466 passed / 1 failed = the documented Isolate bootstrap payload test from real settings #684 bootstrap env baseline (dev machine only; passes on CI hosts).
• cargo clippy both crates (shared + tauri) --all-targets -- -D warnings — 0 errors at 1.98.0.
• Frontend: vitest 437/437 pass (incl. the new MenuCardOpenAIUsage, OpenAIApiUsageChart, BarChart, chartGeometry, useChartAnimation, IdentitySection tests); pnpm run check-locale OK — 892 keys; pnpm run lint 0 errors; pnpm run build pass.
• UI proof: the prior CUA proof at this exact head (813458d) exists — comment issuecomment-5912754275 (tray chart section, Cost/Tokens tabs, 30 daily bars, seeded data, no personal data). Since the head is unchanged, that proof remains at-head; a fresh browser-use pass was additionally run for the OpenAI chart surface during this validation round (see the proof comment below at the same head).
• No push needed (head 813458d already on origin); no code changes required.

Stack note: this branch still contains the OLD #699 head (5d99354) below its merge base — after #699's f328c66 lands, merge the new base in so the flake fix is behind this head too.