TiboTattle 0.1.29

TiboTattle 0.1.29

TiboTattle recognizes GPT-6.1 Terra as a distinct model in local usage,
API-equivalent accounting and model performance. Prices use OpenAI's
October 26, 2026 release date
and official model pricing,
reviewed on that date. Standard prices per million tokens are $1 for input,
$0.05 for cached input, $1.25 for cache writes and $5 for output. Batch and
Flex use half the Standard rates; Fast uses twice the Standard rates. Requests
above 272,000 input tokens use twice the input and cache rates and 1.5 times
the output rate for the whole request. Exactly 272,000 tokens stays in the
short-context band.

Cache-write comparisons retain TiboTattle's existing 30-minute retention
assumption. Twelve-hour retention and regional premiums are not represented.
Terra Ultrafast remains unpriced, and no Terra-specific Codex allowance weight
is inferred from its API prices. Existing models keep their historical price
cards and identities; earlier published model rosters do not acquire Terra
retroactively. Community views can name the model while leaving an unavailable
estimate explicit.

Codex's Academic Researchers and Academic Researchers Plus plans are now
recognized by their official names,
verified against the installed contract on October 26. Their allowance
capacities and multipliers remain unknown, and they do not enter the
community's personal-plan normalization. Older contribution contracts retain
their original plan vocabulary.

A detailed refresh reprocesses available Codex logs to recognize Terra in
retained usage and timing records. Replay preserves token totals and event
identities. Previously stored unknown records cannot be re-identified when
their source log has disappeared; the app keeps that uncertainty instead of
inventing a model.

Model performance defaults to the system time zone and offers a saved UTC
toggle. Daily and weekly charts, labels and hover targets follow the selected
calendar, including daylight-saving transitions and fractional offsets.
Cached reports stay tied to their time zone, so a UTC summary is not relabeled
as a local-calendar result. Exact rolling report bounds and hosted community
UTC publication retain their existing meaning.

An initial speed setting recorded after a turn starts can now qualify its
performance samples when attributable evidence establishes that it preceded
all output. Genuine mid-turn changes, conflicting or ambiguous attribution,
and unreadable evidence remain excluded from a confident speed assignment.
Reprocessing available history repairs classification without changing the
underlying timing measurements. Thanks to
Siilent-Sea for the calendar and initial-speed
work in PR #299, preserved
and strengthened in PR #300.

Large local export verification avoids repeatedly validating the same source
metadata. Claude exports skip oversized transcript records that cannot contain
usage, while retaining limits for records that may contain usage. Quota
analysis also avoids repeated scans and sorts while preserving its calculations.

The desktop runtime advances from Electron 43.5.0 to 43.8.1. Dependency fixes
also update affected URL parsing, HTTP, and hosted test/build components,
including fast-uri, Undici, Vitest, Vite and Zod. The macOS minimum remains
macOS 14.

The client and server share the new model and price contracts, but installing
the client alone does not deploy the hosted service, recalculate community
results or repair previously accepted hosted records. Those steps require the
separate server rollout and its verification. Platform qualification remains
artifact-specific; see platform support.

TiboTattle 0.1.28

TiboTattle 0.1.28

TiboTattle now recognizes GPT-6.1 Sol as a distinct model and includes its
supported Standard, Batch, Flex, and Fast API-equivalent prices for both context
bands. Observed Ultrafast usage stays separate from Fast in usage and model
performance views. Ultrafast prices are available for the supported Astra
combinations; unsupported model and speed combinations remain explicitly
unpriced. The dashboard explains unavailable rates and distinguishes API pricing
from the allowance included in a subscription.

After a detailed refresh, discoverable older Sol 6.1 logs are reprocessed to
correct model identification and supported pricing while preserving token
totals. Records whose model was stored as unknown cannot be re-identified when
the original log is no longer available. Historical model classification can be repaired locally. Resynchronizing
previously accepted records uses the deployed additive correction contract
and remains pending end-to-end qualification.

Plan labels and expected allowance comparisons now use Pro 10× and Pro Max
25×. The configured expectations place Pro Max 25× at 2.5 times Pro 10×,
Pro 5× at half Pro 10×, and Plus at one tenth Pro 10×. These are product
expectations, not measured provider capacity. The existing Pro identity and
history are preserved, and measured community estimates still depend on
available evidence.

When a log explicitly identifies auto-review, usage from October 6, 2026
(UTC) now appears separately from the main Codex allowance. The marked $0.00
is its main-allowance charge; recorded tokens and API-price evidence remain
intact. Earlier usage keeps its original treatment, and mixed periods show
separate rows. Spark keeps its own measured allowance and timeline.

Detailed refresh reparses available logs for this classification while retaining
older evidence whose source has disappeared. Upgrades preserve event identities
across replay and restart, avoiding duplicate usage. Missing historical metadata
remains unknown.

Accountless performance sharing rechecks its protected sharing preference and
current hosted grant after restart or preference changes before preparing or
sending data. A current grant can resume an earlier authorization or
response-version rejection while preserving acknowledged upload progress.
Ordinary polling cannot retry a terminal rejection, and recorded opt-outs remain
effective. Hosted performance publication still requires separate migration,
activation, and qualification.

A temporary overlap between startup sharing checks no longer permanently
pauses usage sharing. The client retries once the protected preference is
available, while keeping saved opt-outs, current authorization and duplicate
upload protection.

The shared dashboard removes obsolete account, profile, and social controls
while retaining Electron desktop controls, including sharing an image. Project
token mix now shows separate input and output bars, making cached input and
reasoning output easier to compare. Model performance retains every period and
speed view, fixing a case where the default seven-day view stayed on Updating.

Large local review exports now support the runtime’s existing source-file limit,
while preserving older export formats, verification, deletion and recovery.

The desktop runtime updates Electron from 43.2.0 in the published 0.1.26 source
to 43.5.0, with affected transitive dependencies refreshed for security
advisories. The macOS support floor remains macOS 14.

Build tooling removes obsolete download dependencies and retires their temporary
advisory exception. Download integrity, secure transport, cancellation and
platform signing checks remain enforced.

TiboTattle 0.1.26

TiboTattle 0.1.26

Version 0.1.25 was tagged but not published. This release carries its changes
forward. It restores direct Codex allowance reads when recent ChatGPT or Codex
desktop builds place their CLI at Contents/Resources/codex-cli/bin/codex.
TiboTattle also checks the older bundled location, system and user Applications
folders, an explicit CODEX_BIN setting, and then PATH. The doctor identifies
the selected source and bundle layout without displaying a private filesystem
path. Local usage analysis remains independent of the CLI; if a direct quota
read cannot run, the previous observation remains stale. Fresh local usage no
longer relabels that old quota as live. After a terminal refresh timeout, the
dashboard gives cleanup a short grace period before reporting that analysis
did not finish.

On macOS, a locked Keychain can report the same authorization error as an item
access denial. When exactly one matching app credential belongs to a locked
Keychain, TiboTattle now presents the unlock-and-retry guidance. Other access
failures retain the permission-denied guidance. The check is silent and does
not read or replace a credential, change its access policy, or modify the
Keychain search list.

Since the published 0.1.24 release, the desktop source also adds a macOS
update-available indicator, repairs tray cleanup during updater shutdown,
persists startup checkpoints before early quits, warms cached dashboard
reports across periods and speed modes, and corrects the Trends header range
on cached plan views. The model catalog includes newly recognized Claude
names. The forecast-only GPT-6.1 Astra identity remains explicitly unpriced
in local Codex usage reports.

The macOS download requires macOS 14 or later. Each platform's installer and
its verification evidence are listed in the release manifest.

TiboTattle 0.1.24

TiboTattle 0.1.24

GPT-6 Sol and Luna have distinct client and server API-price-equivalent rates
for Standard, Batch, Flex and Fast processing (API priority pricing), including
cache reads, cache writes and long context above 272,000 input tokens. They
reuse their GPT-5.6 family icons while preserving older models and historical
pricing.

Model performance becomes available in Windows x64 Electron. Tool-free turn
throughput complements output speed, while cached performance and work reports
remain visible during refresh.

Projects & threads repairs Windows saved task-name lookup and project grouping,
including Windows Git output and saved repository origins when Git is unavailable.
Windows also reads Codex's selected task-history heads, preserving valid forked
tasks when a paginated history has multiple physical files. macOS and Windows
share the same metadata parity tests and avoid creating source sidecars when
reading a closed database.

Automatic refresh recovery is more resilient, and start-at-login status updates
without reopening Settings. The tray presents clearer pace and run-out timing.
Local opt-in macOS crash capture, content-free startup diagnostics and an offline
support doctor make startup failures easier to investigate.

History ingestion now distinguishes accounting records from nested session
content and accepts rollout lines up to 512 KiB. Parser recovery preserves
previously indexed usage and avoids adding it again on restart.

Optional sharing recovers from more temporary service failures and schedules
an upload after a successful index update. Local diagnostics distinguish refresh
progress from upload progress. Durable opt-outs and retry backoff remain in place.

Model performance keeps Standard and Fast results separate, defaults
to Standard, and counts historical measurements with unknown mode as excluded.
Sparse Fast samples retain their actual counts.

Telemetry v1.2 adds qualified turn-boundary and continuity evidence for optional
sharing while preserving older-client compatibility and existing cache calculations.

TiboTattle 0.1.23

TiboTattle 0.1.23

Remaining allowances become animated fuel tanks with one shared pace forecast,
reset countdowns and estimated run-out timing. Waiting states explain when more
fresh evidence is needed, and stale readings never imply active consumption.

Trends brings allowance, spending, quota comparison and cumulative drift into
linked charts with typed reset evidence. Allowance Value adds plan-specific
five-hour history alongside seven-day estimates.

The cache continuity matrix shows all time buckets together, with sample counts,
percentage-first inspection and model filtering. Shared dashboard controls and
loading states preserve the last matching results during refresh.

A local usage coach and Codex plugin can explain recorded usage through a
bounded, content-free protocol.

Windows file identities and collector/model-timing recovery are strengthened.
Windows model performance remains unavailable while its separate protected
reader qualification is open. This release preserves that feature boundary.

TiboTattle 0.1.22

TiboTattle 0.1.22

Projects & threads shows where your recorded tokens and API-price equivalent
go, with project expansion, thread details, model filters and live local search.
Shared work stays grouped with its parent task without double counting.

Model performance adds output-speed and first-token-latency charts, model and
date selectors, linked hover readouts and measurement details. These are local
timing estimates, with missing evidence kept visible. This page is available on
macOS and Linux; Windows timing support is not available yet. Other local usage
analysis remains available on Windows.

Automatic contributions can prepare retained histories with more than one
million quota observations without repeatedly failing before the first upload.
Large uploads resume validation across bounded passes when their evidence is
unchanged, while account and source changes still trigger revalidation.
Existing installation credentials, sharing choices and opt-outs are preserved.

The refresh toolbar keeps its phase, count and timer in fixed positions.
Allowance-history headings and controls wrap when space is tight. Existing
model icons, including the fallback icon for unrecognized models, are retained.

Completed history scans no longer claim that older sources are still being
indexed. Cache-drop and switch-overhead rows resolve local thread titles from
their own verified history generation, and withdraw links when a match becomes
ambiguous.

Accounting summaries accept the provenance written by the current history
scanner, repairing a compatibility mismatch that could leave a finished scan
without an updated summary.

Model performance and Projects & threads keep their last matching results
visible during background refresh. Changing a period or filter preserves the
correct scope, and unavailable results remain clearly identified.

TiboTattle 0.1.21

TiboTattle 0.1.21

This release completes the update path from the native Mac app to the unified
desktop app. Native 0.1.18 users can choose Check for Updates, install the
update and continue with their local history, settings and contribution choice.
Manual installation also preserves existing local data; no backup-folder setup
is needed.

Both Mac installers now declare the supported minimum of macOS 14 correctly,
fixing the metadata mismatch that caused Homebrew's cask audit to reject them.
The Mac app retains the public verification key required by the native updater
and uses Electron's updater for subsequent releases.

Downloads are available for Apple Silicon, Intel Macs, Windows and Linux.
The website's download buttons use GitHub release assets and provide checksums.
The Mac installers are Developer ID signed and Apple notarized; the Windows
installer is Authenticode signed and timestamped.

TiboTattle 0.1.20

TiboTattle 0.1.20

Install the new Mac app and continue with your existing history and settings. You no longer need to create handover folders, save a copy of the old application, or update native 0.1.16 first.

The app automatically detects compatible retained native data, makes a backup, and carries over history, identity, sharing choices, language, appearance, refresh and tray preferences, and selected activity folders. It also handles installations where the old application has already been replaced. Reopening the app does not repeat the import or duplicate records.

Existing Electron users receive the update through the app's normal updater. Installers remain available for Apple silicon and Intel Macs, Windows x64 and Linux x86_64. Native Sparkle feeds remain separate.

This patch does not require a social sign-in or change sharing choices. Damaged or incompatible retained data produces a recovery message rather than being replaced with an empty history.

Source: f518126a05a6d165b9617f6417a87219d7047273, build 2026091104. Changes since 0.1.19. The attached manifest and verification guide identify final artifact hashes and the platform-specific validation boundaries.

TiboTattle 0.1.19

TiboTattle 0.1.19

TiboTattle moves to a shared Electron desktop app for Apple silicon and Intel Macs, Windows, and Linux.

A more useful desktop companion

• Customize the menu-bar or tray display and open allowance details without leaving your work.
• Get desktop notifications and automatic updates. When a refresh is overdue, retained tray values show that they are stale.
• Read clearer usage and overhead summaries, with recognizable model icons and improved allowance history. Cache-detail lists now say when they show only the most recent records.

Sharing without a social account

Fresh Electron installations explain automatic sharing at first launch. Content-free measurements can be contributed without signing in to Apple or Google; prompts, responses and raw session content are excluded. Analysis remains local.

You can keep sharing off in Settings. Existing explicit sharing choices are preserved. Existing installations that have never chosen receive three visible notices before delayed automatic activation; choosing Keep sharing off stops that transition. An installation is not counted as a unique person. Turning sharing off stops future uploads; it does not erase data already accepted by the service.

Moving from the native Mac app

Preserve the old app before replacing it. Follow the website's guided handover for native 0.1.17 or 0.1.18 so the new app can carry over your history and settings. Native 0.1.16 users should first use Check for Updates in the old app to reach native 0.1.18. If that update is unavailable, or migration is blocked, keep the old app and data and contact support.

Existing native downloads remain available. Their Sparkle updates do not automatically replace the native app with Electron.

Downloads and verification

Choose the installer for your system: macOS 14 or later on Apple silicon or Intel, Windows 10 or later on x64, or the Linux x86_64 AppImage. AppImage mounting requires FUSE 2 support.

Mac installers are Developer ID signed and Apple notarized; the Windows installer is signed. Linux is supplied with checksum verification. See release-manifest.json, SHA256SUMS, SHA256SUMS-ALL-RELEASE-FILES and verify-release.md for exact bytes, platform evidence and any explicit acceptance limits. Checksums and signing evidence do not imply that every physical platform has been exercised.

App source: 178315c4, build 2026091002. Verification tooling is pinned separately in verify-release.md; it does not change the app's source or version tag.

Changes since 0.1.18

TiboTattle 0.1.18

TiboTattle 0.1.18

TiboTattle 0.1.18 adds GPT-6 Astra accounting, fixes paginated Codex history
handling, and introduces a separate Intel Mac installer alongside Apple silicon.
Both require macOS 14 or later. The stable build is 1026; the final release
manifest identifies each architecture's exact installer and verification evidence.

Provenance: GitHub release ·
annotated source tag ·
changes since v0.1.17.

Codex and GPT-6 Astra compatibility

The app recognizes GPT-6 Astra and other reviewed OpenAI models throughout
local accounting and the admin model dashboard. Astra pricing covers Standard,
Batch, Flex and Fast, including cache writes and the request-specific long-context
boundary. Missing token components remain unavailable instead of becoming zero.

Changing Astra's reasoning effort is no longer automatically classified as a
cache reset: cache analysis uses model-specific effort semantics and measured
cache behavior rather than assuming continuity or loss. Current Codex
compaction headers, including their new ordinal field,
are recognized so compaction remains part of cache-drop attribution. Read-only,
bounded Zstd support restores coverage for compressed Codex histories on runtimes
with native Zstd support.

Resumable metadata exports now accept paginated histories that start at ordinal
zero with no physical history base. Logical forks remain independent, repeated
physical generations retain distinct occurrence identities, and copied legacy
tool calls are suppressed consistently. Frozen exports keep their original
parent selection across resume. Older incompatible export workspaces are
refused before mutation; exact physical-base continuations remain unsupported
in the checkpoint export lane rather than being approximated as resets.

Paginated accounting also keeps inherited settings at the correct history
boundary. A reset without model or speed evidence stays unknown; it cannot
borrow the parent's later settings during reanalysis and change historical
pricing attribution. Explicit physical-base continuations retain their pinned
settings, including unknowns, consistently across rebuild and resume.
Descendants use the selected history rather than retired exports, preserving
real usage after a reset without borrowing discarded settings or snapshots.

The admin model view includes all 39 reviewed OpenAI identities, supports
filtering, preserves older history and keeps Spark's allowance separate.
Response-level usage records continue to reconcile through the existing legacy
stream; response-only histories and applied reasoning updates remain unqualified.
API-equivalent pricing remains a comparison measure, not a provider bill or a
claim about subscription allowance formulas. Installing the desktop release
does not apply hosted migrations or activate the hosted admin view.

Separate macOS installers

This is the first release with separate Apple silicon (arm64) and Intel
(x64) installers for macOS 14 or later, built from one release source. It addresses
issue #93 without changing
the already published 0.1.17 assets or tag.

The Intel build uses the native launcher, bundled Intel Node runtime and native
Keychain broker. Its updater uses a separate feed; existing Apple silicon
update URLs stay unchanged. The website has separate macOS tabs and exposes
Intel downloads only when their own validated release evidence is supplied.

Stable build 1026 follows private RC3 1025.2; earlier RC1 1025 and RC2
1025.1 artifacts remain unchanged. Private candidates are not relabeled as
stable, and their checks do not substitute for final stable-artifact evidence.

The release corrects the first-upgrade refresh deadline: reviewed published parser
v10/v11/v12/v13-to-v14 rescans receive the bounded four-hour cold-refresh window,
so a large reanalysis is not interrupted by the ordinary five-minute limit.
Warm refresh deadlines and compatibility safeguards remain unchanged.

Qualification boundary

For this release only, the owner accepted the missing disposable-profile
manual Login Item/lifecycle matrix and physical Intel qualification, and
authorized both Mac installers to proceed to publication. The owner reports
other testers are running the app successfully; this is user-reported feedback,
not a retained architecture-, source- or artifact-bound qualification receipt.
Cross-compilation and Rosetta do not prove physical Intel behavior. The exact
scope and retained safeguards are recorded in the
0.1.18 manual-qualification waiver.

The waiver does not waive data-loss or integrity failures, final signing and
notarization, updater verification, or unexpected Keychain security prompts.
Do not bypass macOS protections or reopen upgraded state with an older app.
No Windows, Linux, Electron or new Homebrew Intel support is claimed. Hosted
migrations, device pairing and contribution activation remain separate from
the desktop release and retain their consent requirements.

Follow the Intel implementation plan
and macOS release runbook.