feat: Codex beta provider with API-equivalent cost (v2.3.0) by Carl723000 · Pull Request #93 · ClaudeCodeUsage/ClaudeCodeUsage

Carl723000 · 2026-08-24T11:28:15Z

Summary

Adds Codex as a beta provider on the shared dashboard, with the same overview, time-range, session, project, settings, and responsive UI foundations used by Claude.
Adds API-equivalent cost and weekly allowance-value estimates from local token logs.
Merges usage from multiple sign-ins found in one Codex home while keeping quota observations conservative.
Keeps verified usage visible, with progress and explicit partial-state wording, while a large Codex index converges.
Accelerates one-time Codex backfills with a bounded adaptive local worker pool, then returns completed indexes to the existing low-power incremental path.
Fixes [Bug]: v2.2.2 — one changed file re-reads the entire 1.1 GB corpus; Extension Host pegged at 100% CPU / 3.9 GB RSS #87 by moving Claude runtime refresh onto an exact in-memory per-file index; unchanged refreshes read zero JSONL bodies and changed files no longer trigger full-corpus aggregation.
Resolves Claude quota credentials from the active profile and locks the established transcript deduplication semantics with regression fixtures.

Metric semantics and limits

Request-level Codex attribution prefers the input, cached-input, output, and reasoning components in a valid last_token_usage snapshot. Its total_tokens is active-context size, not usage attributed to that request.
A full numeric total_token_usage plus last_token_usage signature suppresses only proven replay from the same pseudonymous rate-limit source or an immediately adjacent duplicate. The persisted replay map is bounded; raw account or rate-limit identifiers are not stored.
If last_token_usage is absent, cumulative total_token_usage remains a compatibility fallback and is reduced through the existing component-wise lineage high-water rules.
Existing fork/lineage reconciliation remains the ownership layer. Parallel per-file passes publish results in deterministic order, and regression tests require the complete parallel index DTO to equal the serial result.
The parser-state upgrade triggers one fail-closed reindex. Pre-upgrade contributions stay out of displayed usage until rebuilt, while completed files and their verified subtotal survive partial checkpoints and extension restarts.
API-equivalent cost applies current official API token prices to locally recorded usage. It is an estimate, not a bill, subscription charge, or official quota value.
Unknown Codex models remain unpriced. Pricing coverage uses all processed tokens in the selected scope as its denominator.
Reasoning tokens are a subset of output tokens and are never charged twice. A priced cost card displays $…; a scope with no priceable tokens displays — rather than a manufactured zero.
During incomplete indexing, the dashboard shows only the verified indexed subtotal and current progress; unverified legacy totals are excluded.
Stable duplicate-session identity ambiguity retains both local copies and a visible quality warning, but no longer falsely leaves an otherwise complete index labelled as still running.
Usage from multiple logins in the same Codex home is combined. Quota cards show the last observation and are not summed because Codex logs do not provide a reliable account identity.
Codex remains beta. Branches, Workflows, Cache Warmth, and content attribution remain Claude-only.

Indexing performance and runtime boundaries

A first or incomplete Codex backfill uses up to half the available logical CPUs, capped at six local file-pass workers. Main, lineage, period, and identity passes remain independently bounded and checkpointed.
Live progress remains frequent. Large durable snapshots are coalesced to roughly every 10 seconds, 2 GiB, or 256 completed passes, plus stage boundaries; a small warm append is saved exactly once.
Global lineage is reconciled only in stages that can change it. Completed indexes dispose the extra workers and return to the 64-pass / 128-MiB automatic budget; unchanged refreshes read no usage JSONL body and write no index.
On the maintainer's 16-logical-CPU machine, a read-only cold benchmark rebuilt 2,117 files / 20.79 GB to complete coverage in 66.7 seconds, including 34.75 GB of main and lineage pass I/O and 21 durable checkpoints. This is corpus- and hardware-specific evidence, not a universal performance guarantee.
Claude's production index is in memory for v2.3.0: one Extension Host cold-build remains, but subsequent append/truncate/replace/move/delete updates touch only affected files and aggregate groups. On a 681-file / 352.7-MB local corpus, cold build took 3.34 seconds; an unchanged incremental update took 8 ms with zero body reads, parsed lines, and aggregate mutations.

Claude-side changes

Fixes [Bug]: v2.2.2 — one changed file re-reads the entire 1.1 GB corpus; Extension Host pegged at 100% CPU / 3.9 GB RSS #87 with the production in-memory per-file index while preserving the established full-loader results for response identity, request-ID degradation, content UUID ownership, titles, context, sessions, projects, branches, workflows, costliest messages, and time buckets.
Fixes Quota gauge always reads ~/.claude/.credentials.json, ignoring CLAUDE_CONFIG_DIR and dataDirectory #89 by resolving credentials from the active Claude profile. Custom profiles never fall back to the default macOS Keychain credential; the Keychain fallback remains available only for the default profile.
Locks response-identity deduplication for thinking/text transcript pairs, monotonic snapshots, cross-file clones, distinct request IDs, and mixed request-ID presence.
Changes two visible shared-dashboard details: active-tab text colours and the Cost Composition segment colours.
Preserves expanded rows, sorting, chart metric selection, and scroll position across webview reloads.

Attribution

Contributor attribution appears only once: zeyutang's feat: official quota thresholds and a tidier usage credits tooltip row #90 is already part of upstream main, so this PR does not duplicate that contribution.

Deferred work

Accepted for v2.3.1: the German header clipping near a 980 px viewport.
Accepted for v2.3.1: corrupt-index quarantine cleanup, including stale *.lock.pending-* directory reclamation.
Accepted for v2.3.1: provider-aware model-name variants.
Automatic attribution of $20/$100/$200 subscription tiers remains deferred because local Codex logs do not expose a reliable account-plus-plan identity. A future implementation must be explicit opt-in rather than guessed.
Provider-specific visual adaptation remains follow-up work; this PR keeps Codex inside the established shared-dashboard design system.

Validation

npm run test:release
- Node: 588 tests — 587 passed, 0 failed, 1 capability-gated skip
- Playwright: 53/53 passed
The token-semantics fixtures cover exact last usage, interleaved sources, same-source replay, cross-source reset reuse, last-only and missing-last records, cache aliases, counter regressions, persisted bounded replay evidence, and interrupted one-time reindex resumption.
New performance-semantic fixtures cover parallel/serial deep equality, worker cancellation and disposal, resumable checkpoint cadence, no duplicate final save, stable ambiguity wording, Claude full-loader equivalence, append tails, incomplete tails, request-ID cardinality changes, truncate, atomic replacement, move, delete, content UUID ownership, failed-read atomicity, and bounded 100-file updates.
Final local candidate: claude-code-usage-2.3.0-codex-beta-r10.vsix, 82 files, 872,284 bytes, SHA-256 1215c1c31584908cbda7d41d722bd473ef19d0f68315bf8b4214f02a563103fa. Package policy, ZIP integrity, and isolated macOS VS Code installation all passed; the isolated install reports growthjack.claude-code-usage@2.3.0.
package.json remains at 2.1.1; Release Drafter is intentionally driven by the maintainer-approved feature and minor PR labels for the v2.3.0 minor release.

🤖 Generated with OpenAI Codex

Capture the approved provider experience, incremental indexing, local optimization guidance, announcement cleanup, and release acceptance criteria in bilingual review documents. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

Translate the approved design into three self-reviewed, TDD-oriented implementation plans covering release hygiene, provider/index architecture, and the Codex-specific product experience. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

Add a pure, tested full-version selector so upgrades never fall back to stale major/minor release notes and fresh installs stay quiet. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

Wire exact-version v2.3.0 release notes into the extension, add a default-on opt-out setting, keep fresh installs quiet, and localize the announcement across all eight UI languages. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

Update the bilingual agent policy and architecture for the approved v2.3.0 provider boundaries, record the released v2.2.1 date, and preserve the existing automation hardening assertion. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

Add explicit provider token, confidence, outcome, event, and limit contracts with tested processed and fresh-input semantics that prevent cached and reasoning subsets from being double-counted. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

Add a golden-tested compatibility adapter that wraps the existing Claude all-time aggregation without changing token or cost results. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

Add minimal schema guards and a synthetic-fixture-tested high-water parser for cumulative Codex token snapshots, turn context, subset semantics, quality flags, and local last-observed limits. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

Exclude inherited parent counters, flag regressions without negative usage, pseudonymize session/project lineage metadata, distinguish approval reviewers, and retain only structural tool facts. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

Codex discovery is restricted to session JSONL allowlists, keeps raw paths in memory only, and differentiates append, truncate, replacement, move, and deletion for incremental indexing. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Store only privacy-safe per-file aggregates, tail unchanged logs from the last byte offset, reuse archived sessions, and retain verified totals when one file fails. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Run local manifest discovery and incremental parsing in a cancellable single-flight worker, with privacy-safe progress, results, and errors. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Add the extension-facing Codex provider lifecycle, preserve the last verified snapshot on failure, and emit only anonymous index diagnostics. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Add safe default-on local Codex aggregation, a 30-second quiet watcher option, explicit status-bar provider and metric choices, and localized settings across all UI locales. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Aggregate recent-task, 7-day, 30-day, and project views with Codex-native processed, fresh, cache, reasoning, model, effort, lineage, coverage, and quality semantics. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Add deterministic local guidance for child-thread fresh usage, high-effort small changes, repeated post-change command structure, cache context, and approval-reviewer overhead. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Render Codex-native usage scopes, dimensions, coverage, last-observed limits, local optimization cards, and a side-by-side provider comparison without cost or quota equivalence. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Add provider-aware dashboard selection, Codex-only startup support, side-by-side comparison, interactive Codex scopes, and complete localized provider copy across all eight UI locales. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Create the local Codex provider without eager scanning, refresh it in a background worker, watch only allowlisted log roots with a 30-second quiet debounce, and switch status metrics without losing Claude state. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Document provider-specific metrics, local privacy boundaries, watcher controls, and release behavior across all seven README editions while preserving the original PR #38 credit. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Treat child counters as independent, preserve lineage across repeated metadata, map guardian sessions to approval reviewers, and ignore known non-usage envelopes. Clarify the local-content privacy boundary in every README. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Record the verified test suite, final VSIX audit, isolated activation smoke test, privacy-safe 2.58 GB benchmark, beta boundaries, and maintainer release path in Chinese. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Define the provider-native modules, shared visual language, behavior diagnostics, settings integration, and R2 verification plan. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Add detailed thread and daily rows, all-time and monthly scopes, token composition, and provider-native behavior metrics with focused tests. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Add eight provider-native modules, rolling and monthly chart data, Claude-style bar and composition visuals, compact formatter hooks, and behavior summaries. --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/) Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

Replace the parallel Codex render stack with provider-aware dashboard functions, rebuild Sessions as eight columns with expandable details, and retire obsolete Codex-only UI files and snapshots. Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Use one uncached terminology family across all eight locales and settings copy while preserving internal field names and persisted values. Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Restore expanded rows, table sorting, chart metrics, active destinations, and scroll position through the VS Code webview state API. Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Replace semitransparent chart aliases used as solid fills, strengthen active and sortable text cues, and remove production DEBUG console output without suppressing real errors. Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Document the Codex beta shared dashboard, uncached terminology, privacy and cost boundaries, faithful UI harness rules, and the distinct product and development-tool roles. Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Rebuild token lineage from ordered numeric fingerprints, remove only verified copied prefixes, preserve independent siblings, migrate legacy indexes safely, and expose incomplete backfill progress and missing-parent quality in the shared dashboard. Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Document the numeric lineage contract, conservative missing-parent behavior, active/archive overlap handling, legacy-index rebuild, and the user-visible incomplete-index progress state in English and Chinese. Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Retain the v2.3.1 benchmark and fixture sources while explicitly excluding the prototype and benchmark artifacts from the published VSIX. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Merge multi-sign-in usage while retaining last-observed limits, publish persisted verified subtotals during index convergence, and add API-equivalent cost and weekly value estimates. Keep Claude quota credentials aligned with the active profile, including the default-profile-only Keychain boundary (#89). Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Cover copied replay prefixes, post-fork counter resets, independent identical siblings, nested forks, and persisted replayed appends with exact last-usage snapshots. Preserve the existing cumulative-fallback fixtures and runtime behavior. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Remove insertion-order sensitivity from the existing cumulative scan-parity test and the new exact-last test. Compare complete payloads by file key and explicitly verify a reversed file map, preserving all token and lineage assertions. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Use a bounded adaptive worker pool, coarser resumable checkpoints, and stage-aware lineage reconciliation for one-time Codex backfills while preserving the low-power steady-state path. Move Claude runtime refreshes onto an exact in-memory per-file incremental index so changed files no longer trigger full-corpus body reads or full-record aggregation. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Unify reset-aligned periods, withhold ambiguous allowance inference, add Codex cache-hit parity, and make the weekly value panel a dashboard-only setting without changing the index schema. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Use the localized Today label without changing recent-task semantics, and keep rolling 30-day charts and tables inside reliable horizontal scrollers. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Give the rolling 30-day table an explicit readable track width so Linux and macOS Chromium both expose the local horizontal scroller. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Make Today a real calendar-day scope with a targeted resumable hourly sidecar, default daily and monthly charts to API-equivalent cost, preserve unknown-model pricing boundaries, and align shared Claude/Codex time-series layouts. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Carl723000 and others added 30 commits August 21, 2026 20:28

feat: integrate Codex dashboard settings

bcee972

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

feat: localize the Codex dashboard

43d256b

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

docs: update the Codex dashboard handoff

5a6d2d5

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> --- 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

docs: specify truthful Codex identities

420bb14

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

feat: preserve truthful Codex identities

01dd815

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com> 🤖 Generated with [OpenAI Codex](https://developers.openai.com/codex/)

Carl723000 and others added 13 commits August 21, 2026 20:36

test: lock Claude transcript dedup semantics

b052c70

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

chore(#87): keep the incremental prototype out of the package

1e1d243

Retain the v2.3.1 benchmark and fixture sources while explicitly excluding the prototype and benchmark artifacts from the published VSIX. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

test(ui): cover Codex cost and indexing states

38ec9e3

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

docs: record v2.3.0 behaviour and the session_index exception

2c1dcba

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

fix(codex): remove redundant cost approximation marker

7a3f331

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Carl723000 requested a review from jack21 as a code owner August 24, 2026 11:28

Carl723000 added feature Minor bump — new feature minor Minor bump labels Aug 24, 2026

Carl723000 and others added 10 commits August 26, 2026 01:18

fix(codex): attribute exact request token snapshots

35fd9b6

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

docs: record Codex request token semantics

7af32f5

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

test(codex): lock resumable token semantics rebuild

c406b45

Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

fix(codex): keep daily views readable

07d1598

Use the localized Today label without changing recent-task semantics, and keep rolling 30-day charts and tables inside reliable horizontal scrollers. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

fix(codex): stabilize daily table scrolling

9e46634

Give the rolling 30-day table an explicit readable track width so Linux and macOS Chromium both expose the local horizontal scroller. Co-authored-by: OpenAI Codex <215057067+openai-codex[bot]@users.noreply.github.com>

Carl723000 merged commit eca4e43 into main Aug 28, 2026
4 checks passed

Navigation Menu

Repository navigation

feat: Codex beta provider with API-equivalent cost (v2.3.0) - #93

feat: Codex beta provider with API-equivalent cost (v2.3.0)#93
Carl723000 merged 135 commits into
mainfrom
codex/v2.3.0-lineage-r6

Carl723000 commented Aug 24, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

Carl723000 commented Aug 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Metric semantics and limits

Indexing performance and runtime boundaries

Claude-side changes

Attribution

Deferred work

Validation

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Carl723000 commented Aug 24, 2026 •

edited

Loading