Changelog

Pulled live from the repo on each build. Last build: Thu, 21 May 2026 09:38:26 UTC.

Source: CHANGELOG.md on the tokentrace main branch. The page below is a verbatim render, no transformation.

Changelog

All notable changes to TokenTrace are documented here.

[0.14.2] - 2026-05-21

Added

  • Added get_agent_guide to the MCP server with registry install snippets, recommended agent workflows, copy-paste AGENTS.md guidance, and local-first guardrails.
  • Added tokentrace mcp selftest --json to verify MCP initialization, tool listing, agent guide output, and scan-confirmation refusal without scanning files.
  • Added docs/agent-adoption.md with MCP setup instructions and evidence-backed recipes for coding agents.

Changed

  • MCP tool responses now include an agent-oriented envelope with summary, confidence, next actions, warnings, evidence hints, human-confirmation state, and the underlying data payload.
  • Package inspection and CLI smoke checks now verify the MCP agent guide, self-test, and adoption documentation.

Fixed

  • Data-backed CLI command help now prints before touching local runtime state, so tokentrace scan --help, doctor --help, evidence --help, repair --help, and related commands stay safe on broken or fresh local databases.

[0.14.1] - 2026-05-20

Fixed

  • Added the required npm mcpName package metadata so the MCP registry can verify the TokenTrace npm package.

[0.14.0] - 2026-05-20

Added

  • tokentrace mcp now starts a local stdio MCP server for capabilities, status, Scan Health, evidence, repair queue, reports, and explicit local scans.
  • Added a validated MCP registry manifest for the npm package, with tokentrace mcp as the stdio entrypoint and no placeholder secrets.

Changed

  • Agent discovery, package smoke checks, packed-install smoke checks, and package inspection now cover the MCP server surface.

[0.13.0] - 2026-05-20

Changed

  • Overview now uses focused data assembly and smaller panel components, with quieter local development overlays.
  • Overview trend aggregation now avoids SQLite localtime bucketing so large local databases load the first dashboard view much faster.
  • Overview now uses a page-specific analytics profile to avoid computing secondary page insights during first load.
  • Slow dashboard analytics queries now surface through internal timing guardrails and doctor output.
  • CLI startup code is split into focused context, help, runtime, serve, and command modules while keeping the executable entrypoint thin.
  • Analytics queries are split into summary, trends, entities, repair, scan-trust, and insight domains to reduce hotspot risk.
  • Scan Health is split into focused diagnostics panels so the route stays limited to data assembly and composition.
  • Session Explorer now has sticky dense-table headers, clearer active-filter chips, row-density controls, faster reset actions, and stronger empty states.
  • Session Explorer filtering, summaries, high-cost thresholding, active-filter labels, and saved-view serialization now live in pure helper functions.
  • Repair now highlights the top unknown-cost cause, next best repair, expected impact, and clearer resolved, ignored, and parser-review state copy backed by executable repair actions.
  • Repair page data loading, summary cards, guidance, and repair queue table now live in focused modules so the route stays small.
  • Model Rates now has clearer repair-context copy, edit validation, duplicate provider/model warnings, save-result feedback, and CSV import/export for bulk rate changes.
  • Codex CLI ingestion now handles JSON-array session exports, warns on malformed local records, and persistence now reports empty, duplicate, or mismatched replacement imports before writing.
  • Unknown-cost repair actions are split into a focused helper module for clearer repair causes, next actions, and resolved-state copy.
  • Settings is split into focused scan, import profile, guardrail, package trust, export, storage, and custom-folder sections.
  • Mobile navigation now uses a compact command menu plus priority shortcuts, and first-run setup now leads users through local folders, model rates, scan, Scan Health, and review without cloud or sample-data assumptions.
  • Dashboard tables, page actions, and compact metric cards now behave better at narrow mobile widths.
  • tokentrace serve now prints clearer startup progress, help examples, browser-open guidance, and fixed-port conflict recovery steps.
  • CLI smoke checks are split into discovery, JSON/data, runtime watch, and dashboard serve modules so package smoke failures point to the broken surface.
  • Route loading copy now reinforces that TokenTrace is reading local database records without sending telemetry.
  • Repair and Model Rates now use mobile card layouts on narrow screens instead of forcing wide table workflows.
  • Unknown-cost repair, scan ingestion, Evidence Trail, and Pricing Settings internals are split into smaller focused modules to reduce hotspot risk and make future repair work safer.
  • Pricing saves, model-rate refreshes, and scans now report a repair delta showing unknown-cost before/after counts, resolved groups, and the top remaining blocked cause.
  • Settings now defers dense package trust, guardrail, import profile, and export sections while preserving section anchors.
  • Added npm run browser:guard to fail browser smoke runs on console errors, page errors, Next dev overlay issues, blank charts, and severe mobile overflow.

[0.12.1] - 2026-05-19

Fixed

  • Public docs, CLI help, package runtime output, and agent-readable release status now describe shipped TokenTrace behavior only.
  • Future-version roadmap labels no longer appear in public package surfaces.

[0.12.0] - 2026-05-19

Added

  • Local Sources & Trust release bundle covering native local ingestion, evidence packs, scan scheduling, scoped guardrails, import-profile preview, saved reports, operating metadata export, and agent-readable release status.
  • Native structured usage log ingestion for local wrappers and team JSONL/NDJSON logs with session, model, token, and source-cost fields.
  • Native Cursor-style chat/composer export ingestion with local source evidence and no raw prompt storage by default.
  • Source Catalog and Source Coverage in Scan Health so users can distinguish native, profile-assisted, fallback, and unsupported local files.
  • Privacy-safe Evidence Packs exported as JSON or Markdown with totals, confidence drivers, source files, parser notes, model-rate state, repair links, and an explicit redaction manifest.
  • Local scan scheduling settings for manual-only, on-open, hourly, and daily scans, plus scan-history retention and last scheduled scan status.
  • Scoped project/model/tool guardrails with per-guardrail warning thresholds and anomaly notes.
  • Import Profile preview for sampling a local file, checking parser fit, reviewing detected fields, and applying recommended matchers without exposing raw content.
  • Saved report definitions and export endpoints for weekly usage, high-cost sessions, unknown-cost repair, confidence trends, guardrail status, and source coverage in Markdown, JSON, and CSV.
  • Operating metadata export for settings, source catalog, schedules, guardrails, report definitions, and roadmap status without raw usage records.
  • Agent-readable release status with implemented cards, action recipes, evidence paths, verification gates, and release status.

Changed

  • Settings now includes Scan Scheduling, Scoped Guardrails, Import Profile preview, and Local Exports.
  • Evidence pages now offer one-click JSON and Markdown evidence-pack exports.
  • Evidence remains a contextual drill-down from Overview, Sessions, Repair, and exports, with direct-entry guidance for users who open /evidence manually.
  • Scan, setup, guardrail, package-trust, folder, import-profile, and export CTAs now deep-link to the exact Settings section instead of dropping users at the top of Settings.
  • Settings scan feedback now reports files checked, records imported, warnings, errors, recalculated costs, unknown cost, stale support imports, model aliases, and next-step links to Scan Health, Repair, Discovery, and Model Rates.
  • Browser-triggered Scan now responses now return compact warning/error previews with full counts, keeping large duplicate-file scans from shipping megabytes of warning text back to the UI.
  • Settings now has sticky section navigation, and Scan Controls shows a persisted Last scan result from local scan history after reload.
  • Overview now includes a compact Last verified trust strip near the accounting totals for latest scan, package IOC, model-rate coverage, and evidence-pack availability.
  • Evidence pages now include opened-from breadcrumbs, safe return behavior, and explicit Export JSON pack / Export Markdown pack actions.
  • Route transitions now show a subtle top loading bar, and route loading states explain what is happening plus the next useful action.
  • Scan and evidence follow-up actions now consistently use Scan now, Open Scan Health, Open repair, Set model rate, View evidence, and Export pack language.
  • Period filters stay attached to the current page, preserve Evidence context, and keep desktop controls in one compact row while retaining horizontal scrolling for narrow widths.
  • Chart cards now show lightweight loading placeholders during client hydration instead of appearing blank while Recharts initializes.
  • Session Explorer now paginates large local result sets to keep dense session tables responsive.
  • Repair now opens as a capped workbench of the top visible unknown-cost groups, while keeping full summary counts and focused repair links for deep review.
  • Roadmap CLI/API now report the 0.12.0 release contract, implemented cards, action recipes, evidence paths, verification gates, and release status.

Fixed

  • Overview, Evidence, and Settings now avoid full raw-row scans and oversized scan-health payloads on large local databases by using covering indexes, scoped scan-file reads, and a summary session path on Overview.
  • Repair no longer serializes the entire unknown-cost queue on first load and uses a dedicated cost-repair index for large local databases.
  • Scan Health no longer loads all scan-file evidence and full session details just to render summary panels.
  • Overview no longer blocks initial render on due scheduled scans, so opening localhost does not wait for scan work before showing the dashboard.

[0.11.0] - 2026-05-18

Added

  • npm run security:ioc scans project files and local Claude/VS Code hook files for high-signal Mini Shai-Hulud/TanStack supply-chain indicators before release.
  • Tokenizer-backed estimates for recognized OpenAI/Codex and Claude-family model names, with explicit tokenizer estimate and simple estimate confidence labels.
  • Native SQLite history ingestion for local usage databases with usage-shaped tables, source-file evidence, import-profile metadata, and source-provided cost preservation.
  • Data Confidence scoring on Overview, Projects, Sessions, and Session Timeline views, combining token source, cost coverage, parser confidence, and scan freshness.
  • Import Profiles in Settings so users can define safe local wrapper-log matchers without writing parser code.
  • Repair Workbench bulk actions for marking visible unknown-cost groups verified, parser-review, ignored, or reopened.
  • Supply-chain IOC status in Scan Health, backed by the same local security:ioc scanner used by release checks.

Changed

  • Session Timeline pages now explain token spikes, model changes, cache activity, cost coverage, confidence drivers, and direct unknown-cost repair links.
  • Scan Health now distinguishes exact, tokenizer-estimated, simple-estimated, high-confidence, low-confidence, and unknown token rows.
  • The roadmap API and CLI now report the 0.11.0 Accuracy & Evidence release contract.

Fixed

  • Source-provided interaction costs are no longer overwritten as unknown during post-scan cost recalculation when model rates are missing.

[0.10.1] - 2026-05-18

Changed

  • Guide now uses a manual-style layout with section navigation, compact setup status, workflow rows, command tables, and tighter troubleshooting copy.
  • Sidebar Guide access now lives in a Help area above the version footer so it reads as documentation instead of a product feature.
  • Guide section navigation now sticks while scrolling on desktop widths, and the standalone sidebar Guide link no longer shows redundant Help chrome.
  • Guide Scan now controls now run the local scan directly with inline feedback instead of linking to Settings.
  • Overview now groups processed, fresh, and cached token metrics into one Token Accounting card with direct evidence pivots.
  • Evidence pages now include metric tabs and drill-down actions for source files, sessions, parser confidence, and model-rate or repair follow-up.
  • User-facing model price configuration is now labeled Model Rates so it is not mistaken for TokenTrace product pricing.
  • Overview Usage Pulse now labels current, previous, and change values explicitly inside each metric block.
  • Overview now groups cost and sessions into one split card with a shared help tooltip, compact pane labels, and aligned trust notes/actions.
  • Overview trend charts now default all-time views to the latest 30 days, while keeping 60-day, 90-day, and All history options available.
  • Overview now compacts below-chart diagnostics into Review Status and Top repair items strips, with the full unknown-cost table moved to the Repair page.
  • Evidence, repair, parser, and model-rate links now use consistent action labels: View evidence, Open repair, Set model rate, and Review parser.
  • Sidebar navigation now shows the active page, first-run empty states point to the next useful action, direct scans return richer result feedback, and repair/evidence pages guide users through the next drill-down.
  • Page names now stay aligned around Parsers, Discovery, Insights, Scan Health, Model Rates, and privacy-oriented Raw Data copy.
  • User-facing diagnostic copy now uses Scan Health consistently instead of older mixed diagnostic wording.
  • Product metadata, README, Guide, and agent discovery now point to the TokenTrace product website while creator attribution points to Abhi Yoheswaran's homepage.
  • Period filters use a mobile-friendly preset scroller with custom dates on a compact second row, and trend controls now say Display window with a showing-latest badge.
  • README screenshots were refreshed from a guarded public-safe screenshot database seeded by npm run screenshots:seed.

[0.10.0] - 2026-05-18

Added

  • In-app Guide page covering first scan setup, Claude Code status-line installation, status-line label meanings, page workflows, privacy, and troubleshooting.
  • tokentrace agent --json and tokentrace capabilities --json for a read-only machine-readable discovery manifest that coding agents can use before scanning or opening the dashboard.
  • Local dashboard discovery endpoints at /api/agent and /api/capabilities returning the same manifest.
  • tokentrace roadmap --json and /api/roadmap for machine-readable 0.10.0 implementation status, evidence paths, verification gates, and release status.
  • Package-level agent discovery references: TOKENTRACE_AGENT.md, llms.txt, and docs/agent-discovery.schema.json.
  • Repo-level agent instructions for Codex and Claude Code requiring Superpowers, ProjScan, changelog discipline, and explicit maintainer approval before releases.
  • 0.10.0 Guided Operator roadmap for in-app guidance, status-line clarity, trend continuity, release-safe agent workflow, and verification gates.
  • First-run guided setup in Overview and Guide so new users can move from scan roots to first useful evidence without a separate tutorial mode.
  • Guide release-readiness and empty-state sections covering roadmap gates, release status, no data, missing logs, unknown pricing, parser warnings, and sandbox smoke skips.

Changed

  • Guide now shows live local setup status for latest scan, imported records, unknown costs, and priced model coverage.
  • Agent discovery follow-up commands are now structured as command arrays instead of shell strings.
  • Claude Code status-line output now leads with live context and cost, then labels transcript totals as processed and cache usage to avoid confusing cumulative processed tokens with current context size.
  • Package inspection and packed-install smoke now enforce the agent discovery docs, schema, executable CLI bin, and 0.10.0 release-status contract.
  • Overview metric cards now show inline trust annotations explaining processed, non-cache, cached, cost, and session counts near the numbers.
  • Overview Period defaults to All time again, while token and cost trend charts share one control bar that defaults chart history to the latest 60 days with 30-day, 90-day, and All options.
  • Usage Pulse now suppresses extreme percentage labels when the previous comparison baseline is too small to be useful.

Fixed

  • Token and cost trend charts now include explicit zero-value calendar days between imported usage days instead of visually skipping idle periods.

[0.9.0] - 2026-05-13

Added

  • Session Timeline pages for inspecting one imported CLI session as ordered interactions, model changes, token spikes, cache activity, tool calls, parser confidence, and unknown-cost events without exposing raw message bodies.
  • Saved Local Views on Sessions, with built-in local filters for unknown cost, high-cost sessions, Claude/Codex this month, estimated tokens, guardrail review, and parser review, plus user-created SQLite-backed views.
  • tokentrace digest --since <last-scan|yesterday|YYYY-MM-DD> for scoped local usage digests.
  • tokentrace report --markdown and tokentrace review --json for deterministic local summaries covering digest data, post-session movement, accounting state, parser warnings, and expensive sessions.
  • Overview Post-Session Review panel showing latest scan movement, unknown costs, parser follow-up, guardrail state, and expensive sessions.
  • Accounting invariant checks that verify processed tokens balance against fresh input, output, reasoning, and cache buckets.

Changed

  • Claude status line output now includes context-window usage when Claude provides it and distinguishes priced sessions from pricing-repair states.
  • Overview Data Readiness now includes token-bucket balance and keeps readiness tiles readable at normal desktop widths.

[0.8.4] - 2026-05-13

Changed

  • Overview now shows Usage Pulse, top-level metric cards, and token/cost trend graphs before trust and Data Readiness diagnostics, so the first viewport focuses on primary usage status before repair-readiness details.

[0.8.3] - 2026-05-13

Fixed

  • Codex CLI token_count JSON imports now subtract cached input from non-cached input and keep cached input separate, so processed Codex totals count cache exactly once instead of double-counting it.
  • Codex parser provenance is bumped to version 4 so previously imported version-3 Codex session rows are reprocessed on the next scan with corrected token and cost totals.

[0.8.2] - 2026-05-13

Fixed

  • Codex CLI token_count imports now preserve non-cached input tokens and add cached input tokens to processed totals, matching Codex's input (+ cached) session summary.
  • Codex parser provenance is bumped to version 3 so previously imported Codex session files are reprocessed on the next scan instead of keeping stale undercounted rows.
  • Generic text-log parsing now recognizes Codex Token usage: total=... input=... (+ ... cached) output=... summary lines as structured usage instead of weak text estimates.
  • Overview and evidence token totals now switch to billions at large scales instead of displaying multi-thousand million values.
  • Scan, settings, pricing, and pricing-refresh API writes now reject malformed JSON and avoid JavaScript truthiness coercion for boolean flags.
  • Settings and scan custom folders are trimmed and blank entries are discarded before persistence or scan execution.
  • Pricing manifest imports now ignore invalid, boolean, array, blank, and negative price values instead of coercing them into trusted numeric prices.

[0.8.1] - 2026-05-13

Fixed

  • Codex CLI imports now read exact token_count totals, including cached input and reasoning output tokens, so cleared Codex sessions are no longer undercounted.
  • Claude Code and Codex session artifacts can exceed the generic file-size cap without being skipped during discovery.
  • OpenAI-style, Claude-style, and generic usage parsers now normalize cached, cache-write, and reasoning token fields without double-counting them.
  • Parser-version-aware rescans now reprocess stale imports, and source-file replacement is atomic so a failed replacement cannot delete prior trusted sessions.
  • Unknown-cost cause summaries now assign each interaction to one primary cause instead of overlapping buckets.
  • Pricing, scan, settings, repair, and pricing-refresh APIs now reject malformed JSON with clean 400 responses.
  • Manual pricing saves now reject blank model/provider names and invalid numeric prices instead of silently storing unknown prices.
  • CLI commands now reject unknown flags across scan, pricing refresh, status, doctor, digest, and insights commands.
  • Local release checks now run ProjScan through projscan@latest, matching the GitHub security workflow guardrail.

[0.8.0] - 2026-05-12

Added

  • Evidence detail pages and tokentrace evidence --json for tracing metric totals back to sessions, source files, parser status, and pricing context.
  • Unknown Cost Repair workbench and tokentrace repair --json for grouped local repair state, alias hints, parser review links, and pricing follow-up.
  • Parser Trust Report and Scan History Diff panels in Diagnostics for latest scan parser coverage and scan-to-scan import changes.

Changed

  • Overview metric cards now link major totals to evidence trails and route unknown cost work to Unknown Cost Repair.
  • Overview now places Token Trend and Cost Trend directly after Usage Pulse and the metric cards, with Monthly Guardrails and Recommended Next Actions below the charts.
  • Dense evidence, repair, parser trust, and scan diff tables preserve horizontal scrolling and stable source-path truncation.
  • Evidence and repair copy now states local-first behavior, support-file ignores, and parser-review requirements for unsupported files.
  • README screenshots now use public-safe synthetic Evidence + Repair views, and obsolete screenshot assets were removed from the package payload.

Fixed

  • Overview custom period date fields now use an intentionally inset calendar icon while preserving native type="date" submission fields and the single-line desktop toolbar.
  • The app shell now constrains page width on small screens so wide toolbars scroll internally instead of widening the whole page.
  • tokentrace statusline setup claude and piped Claude status-line input no longer touch the TokenTrace app database or start the dashboard.

[0.7.0] - 2026-05-12

Added

  • 0.7.0 Usage Intelligence roadmap for local guardrails, savings review, session comparison, project intelligence, and daily digest work.
  • Local monthly cost and token guardrails stored in Settings.
  • Overview Monthly Guardrails panel showing month-to-date local usage against configured limits.
  • Recommendation rules for monthly guardrails that are near or over limit.
  • Evidence-backed Review Queue on Insights, ranked from guardrails, unknown cost repair, high-impact sessions, dominant projects, model review, and cache reuse.
  • tokentrace insights --json now includes the Review Queue for local automation.
  • tokentrace digest and tokentrace digest --json for current-month local usage, guardrails, top review item, unknown-cost count, top project, and latest scan status.
  • Session Comparison Flags on Sessions to highlight token and cost outliers compared with matching project, tool, and primary-model peers.
  • Project Signals on Projects for dominant usage, unknown cost, estimated-token confidence, and model concentration patterns.

Changed

  • Renamed the Insights page header to Usage Intelligence to match the 0.7.0 product theme.
  • Refreshed README screenshots for Overview, Usage Intelligence, Sessions, Projects, and local guardrail Settings using public-safe synthetic local data.

Fixed

  • Increased the forced packed-install smoke test timeout so native SQLite dependency installation is not killed on slower machines.

[0.6.0] - 2026-05-10

Added

  • 0.6.0 Stable Daily Tool roadmap with explicit release cards and gates.
  • Support matrix for stable, best-effort, ignored, and unsupported TokenTrace surfaces.
  • Scan Doctor support matrix and scan freshness status.
  • Overview first-run checklist that explains missing roots, no scans, zero imports, and next actions.
  • npm run smoke:cli for clean-home CLI checks across scan, serve, doctor, status, Claude status-line, and watch-mode commands.
  • npm run smoke:packed for installing the packed tarball into a temp project and verifying the published CLI entrypoint.

Changed

  • Claude Code and Codex JSONL parsers now keep valid records when a transcript contains malformed lines.
  • Model alias suggestions now handle OpenAI/Codex provider-prefixed and dated snapshot names.
  • Overview period filter keeps the custom date fields and Apply button visible on desktop while presets absorb overflow.
  • Local development config declares localhost/127.0.0.1 dev origins and disables the standard Next.js dev indicator preference.
  • README documents the support matrix and unsupported product boundaries.
  • npm run release:check now includes CLI and packed-install smoke checks before package security inspection.
  • The npm package now excludes generated .next output and prepares the dashboard build in the user's TokenTrace app-data directory on first tokentrace serve.
  • Package trust inspection now verifies the publish tarball directly and fails if generated Next.js build output is included.
  • Packed-install smoke now starts tokentrace serve from the packed tarball and allows dependency install scripts needed by native SQLite bindings while keeping TokenTrace itself free of lifecycle scripts.
  • Dev security posture is cleaner: Vitest is updated to the Node 18-compatible patched line, and the unused drizzle-kit dev dependency has been removed.
  • npm run security:package now audits the full dependency graph at moderate-or-higher severity instead of checking only production high-severity advisories.

Fixed

  • Package inspection no longer depends on the user's global npm cache, which can contain root-owned files on some machines.
  • First-run dashboard preparation from a packed install no longer hits SQLite database is locked errors while Next.js imports API routes during build.
  • tokentrace serve now exits nonzero when the underlying Next.js server fails before readiness.

[0.5.1] - 2026-05-10

Added

  • TokenTrace logo SVG asset and Next.js app icon.

Changed

  • App shell and README now use the TokenTrace logo instead of the generic chart mark.

[0.5.0] - 2026-05-10

Added

  • Package trust inspection that fails if generated Next.js route bundles look packed or unreadable.
  • GitHub Actions Trusted Publishing workflow for provenance-backed npm releases from version tags.
  • GitHub security workflow with package verification, production dependency audit, and ProjScan doctor checks.
  • Release-note extraction for publishing full changelog sections into GitHub Releases.
  • Security policy documenting local-only behavior, package install guarantees, and vulnerability reporting.
  • Package trust documentation for Socket, npm, maintainers, and users.
  • ESLint, Prettier, EditorConfig, and ProjScan configuration for cleaner automated review.
  • npm run lint in the verification path.
  • tokentrace serve --port, --hostname, and --no-open options for predictable local and package smoke tests.
  • 0.5.0 roadmap, design spec, and implementation plan for package trust and usage intelligence.
  • Overview Usage Pulse comparing token, cost, session, and unknown-cost movement against the previous period.
  • Scan Doctor recent scan history.
  • Settings package trust panel summarizing install-script, network, and release guarantees.

Changed

  • Published Next.js server bundles are no longer server-minified so package scanners can inspect generated runtime code.
  • Raised the declared Next.js dependency floor to ^15.5.18 to match the patched locked runtime.
  • Raised the declared drizzle-orm dependency floor to ^0.45.2 and forced patched postcss resolution through npm overrides.
  • Updated release documentation so future npm publishes happen through GitHub Trusted Publishing after tag verification.
  • npm run verify now runs Vitest, TypeScript, and ESLint.

Removed

  • Removed the unused date-fns dependency from the published package graph.

[0.4.0] - 2026-05-09

Added

  • Visible running TokenTrace version in the desktop sidebar, mobile header, and Settings.
  • 0.4.0 roadmap and release checklist for maintainer-approved public releases.
  • tokentrace doctor --json plus a shared Doctor report model for scan health and repair recommendations.
  • Bundled parser provenance metadata on imported scan files, including parser id, display name, source, and version.
  • Unknown-cost repair queue on Overview, grouped by cause, model, tool, source file, and repair path.
  • Deterministic local recommendations on Overview, Doctor, and tokentrace insights --json.
  • Compact and wide live status output modes for tokentrace statusline claude and tokentrace watch --session.
  • Codex integration spike documentation confirming terminal split/watch mode as the supported 0.4.0 fallback.
  • Scan cleanup for previously imported sessions whose source paths are now classified as non-usage Claude/Codex support files.
  • Scan Doctor health now uses the full latest scan-file set instead of the Raw Data table preview limit.
  • Moved the Overview unknown-cost repair queue below the dashboard charts and current mix section.
  • Refreshed README screenshots for Overview, Scan Doctor, Sessions, Discovery, Pricing, and mobile Overview.

Changed

  • Flattened dashboard, Doctor, Settings, Sessions, Insights, and scan-health layouts so nested mini-cards became divider-based sections.
  • Updated dashboard typography with an explicit UI font stack plus monospace numeric treatment for analytics tables.
  • Kept the mobile app header aligned with the desktop trust copy: local only and no telemetry.

Fixed

  • Restored the Overview period filter as a single-line toolbar after custom dates are applied, with horizontal overflow instead of wrapping.
  • Improved badge and button styling so labels and controls remain visible without being detected as card-like nested surfaces.
  • Fixed heading hierarchy across card-based pages by using second-level card headings under page titles.
  • Darkened the primary teal to meet contrast requirements on muted surfaces.
  • Fixed packed global installs serving production builds when next is hoisted outside the package folder.

[0.3.0] - 2026-05-09

Added

  • Claude Code live status-line support through tokentrace statusline claude.
  • tokentrace status --json for machine-readable local usage status.
  • tokentrace watch --session for a terminal-split live status fallback while Codex status-line support remains under review.
  • tokentrace statusline setup claude to print the Claude Code statusLine configuration block.
  • Claude Code status-line README documentation and preview image.
  • Vitest coverage for live status snapshots, Claude status-line stdin rendering, period filter layout, scan discovery, scan health UI, tooltip rendering, and unknown-cost causes.
  • npm run verify as the standard Vitest plus TypeScript verification command.
  • Scan Doctor trust checklist covering pricing, roots, discovered files, imported records, unknown prices, parser warnings, and next action.
  • Ignored non-usage file tracking for local CLI support files.

Changed

  • Tightened Claude and Codex discovery so known support folders and files are ignored instead of parsed as usage.
  • Made generic JSON, JSONL, and log adapters skip known non-usage CLI support paths.
  • Grouped noisy scan notes by reason and examples in Scan Doctor.
  • Broke unknown-cost diagnostics into missing model name, missing pricing, missing token count, and other causes.
  • Expanded parser and provider inference coverage for real-world CLI artifacts.
  • Documented the repo verification flow in README and CONTRIBUTING.

Fixed

  • Kept the overview period selector on one desktop row after custom dates are applied, using horizontal overflow as the fallback.
  • Made metric-card help tooltips visible with an opaque surface and accessible tooltip markup.
  • Avoided treating Claude cache, plugin, and todo files as parser failures.
  • Avoided importing broad Claude support Markdown or Codex support JSON as generic usage.
  • Improved parsing of comma-formatted token counts and numeric timestamp strings.
  • Rejected custom dates that roll over into a different calendar day.

[0.2.1] - 2026-05-09

Added

  • Product and design reference docs for TokenTrace.
  • Impeccable design metadata for local UI review.

Changed

  • Tighten dashboard typesetting with shared page headers, data values, monospace text, and denser table defaults.

Fixed

  • Restore the overview period toolbar layout so preset buttons stay grouped cleanly across desktop widths.

[0.2.0] - 2026-05-09

Added

  • Scan health summary for post-scan trust, including parser review, confidence, pricing coverage, and recommended next actions.
  • Diagnostics CSV export for scanned files and scan runs.

Changed

  • Polish Settings, Discovery, Parser Debug, Sessions, and Overview explainability around scan freshness, confidence, and unknown cost.

[0.1.5] - 2026-05-09

Changed

  • Polish the overview period toolbar and five-card metric layout for denser desktop scanning.

[0.1.4] - 2026-05-09

Changed

  • Clarify overview token metric cards by separating processed tokens, non-cache tokens, and cache read/write tokens.

[0.1.3] - 2026-05-09

Added

  • Overview period controls for all-time, today, rolling windows, this month, and custom date ranges.
  • Date-range-aware overview analytics so headline cards, trends, tool mix, model/project/session aggregates, and insights use the same selected period.

[0.1.2] - 2026-05-09

Fixed

  • Recalculate imported interaction costs after scans, price refreshes, default seeding, and manual pricing edits.
  • Apply configured pricing to duplicate records during force rescans by repricing existing imported interactions.
  • Backfill pricing for observed dated Claude model names, such as claude-haiku-4-5-20251001, from matching base pricing rows.

[0.1.1] - 2026-05-08

Added

  • Bundled public pricing catalog for OpenAI, Anthropic, Google Gemini, xAI, DeepSeek, Mistral, Cohere, and generic fallback models.
  • Pricing refresh support from the public TokenTrace pricing manifest.
  • Cache read and cache write pricing support in cost calculations and the Pricing UI.
  • Provider inference for generic logs so model names such as gemini, grok, deepseek, mistral, and command map to the right pricing catalog.
  • Product screenshots, CLI GIFs, and a more user-focused README.
  • Subtle open-source attribution in the app, README, and npm package metadata.

Changed

  • Preserves manually edited pricing rows when refreshing managed default prices.
  • Keeps public documentation focused on user setup, privacy, screenshots, pricing transparency, and extension points.

Removed

  • Internal publishing notes and source recording files from public documentation/package contents.

[0.1.0] - 2026-05-08

Added

  • Initial local-first TokenTrace dashboard and tokentrace npm CLI.
  • Local filesystem ingestion for AI CLI artifacts with adapter-based parsers.
  • Best-effort adapters for Claude Code, Codex CLI, generic JSONL, generic JSON, and text logs.
  • SQLite storage, migrations, seeding, and reset support.
  • Dashboard pages for overview analytics, tool comparison, models, projects, sessions, optimisation insights, pricing, diagnostics, discovery, parser debugging, and raw data.
  • Token estimation, confidence metadata, cost calculation, CSV export, and duplicate import handling.
  • Optional tokentrace run <command> wrapper mode for local runtime diagnostics.