Repository navigation
Releases
Releases · omrikais/cctally
v1.122.0
Added
- Codex cost reports price
gpt-6.2-astra,gpt-6.2-sol,gpt-6.2-astra-lawandgpt-6-cyberfrom their published rate cards instead of falling back togpt-5rates, including usage recorded before you upgrade. - Claude cost reports price
claude-fable-5-2andclaude-mythos-5-2, including Fast mode, instead of counting them at $0, andstatuslineuses their 1M context window.
Fixed
- The
dashboardno longer briefly drops itsstats.dblocks while checking the database version, which could let the status line write at the same moment a page was reading or a refresh was writing. - After a Codex home is moved or renamed, the dashboard,
codex quotaand the--source codexreports no longer count its sessions, projects, totals and quota history twice until cleanup finishes. cctally codex percent-breakdown --accountnow shows the chosen account's cycle and rejects an unknown account, instead of ignoring the flag.- Codex requests over 272K input tokens on
gpt-6-astra-laware priced at its long-context rates instead of its standard rates. cctally pricing-checkno longer reports GPT-5.6 Sol's promotional pricing as expired before it ends on November 30, 2026.
v1.120.1
Stable upgrade:
v1.118.2→v1.120.1Includes every released CHANGELOG section in this range, newest first.
[1.120.1] - 2026-11-17
Fixed
- With an older SQLite, as many Linux distributions ship, upgrading an old install whose
cache.dbis damaged no longer shows a database migration error; the upgrade waits until the cache is repaired, as it already did with newer SQLite.
[1.120.0] - 2026-11-16
Added
cctally db reindex-searchchecks the conversation search indexes against their messages and rebuilds any that disagree;--checkonly reports.
Changed
- Commands you run now stop waiting on
cache.dborconversations.dbmaintenance after 5 seconds, as they already did forstats.db. Hooks, the status-line writer (record-usage) and background work keep waiting. cctally db statusnow only reads your databases: it no longer resumes an interrupted stats rebuild; the next command that shows or records usage, such asreportor the status line, finishes it instead.- On a large usage history, the
dashboardno longer pauses for most of a second every few seconds while it refreshes, so pages and live updates respond more evenly. It costs about 30 MB more memory. - While
cache.dbis being repaired, thedashboardno longer reads it: conversation search and the reader, Explain and project detail are unavailable until the repair ends, and the Projects panel keeps the data it already shows. - The
dashboardno longer updates or repairs databases while answering a page: a block detail reads stored usage, or your session files during a rebuild, and the dashboard's background work fixes what a page finds. - Right after an upgrade,
db statusanddoctorshow astats.dbfrom an older version as not readable until the next command converts it. - After a gap without a running
cctally dashboard, its transcript catch-up writes about 9% less toconversations.db. Search results are unchanged. - When the Codex hooks that
cctally setupinstalls are enabled, the first Codex turn after deleted Codex sessions are cleaned up no longer redoes that cleanup's quota work, so it stays fast. - With an account selected, the
dashboardConversation Viewer andtranscriptcommands respond much faster: they reuse the conversation summaries they can verify instead of rebuilding the account's whole list. Results are unchanged. - The first cache build after installing, and
cache-sync --rebuild, now write about half as much to disk and finish sooner on large histories. Usage and cost results are unchanged. - Title search in the
dashboardConversation Viewer andtranscript search --kind titleno longer reads every stored message, so it stays fast however large the transcript history is. Results are unchanged.
Fixed
- When
cache.dbexists but cannot be read,cctally pricing-checkno longer says every observed model is priced: it reportsCoverage: unavailable (cache.db could not be read).and a degraded check (cachein--json). - When
cache.dbis found corrupt while an upgrade runs its database steps, cctally no longer deletes it; the next command quarantines a copy and rebuilds it through the normal recovery. - On macOS, remounting the external drive or disk image that holds the data folder no longer makes the next sync rebuild the Codex quota index.
- Codex quota blocks, milestones and alerts stay accurate when updates overlap, Spark windows keep their
dashboardblock detail, and re-enabled alerts notify again. Turning alerts on or changing their rules no longer stalls Codex turns. - After an upgrade, the
dashboardConversation Viewer re-checks its stored session summaries a few at a time, so a session cost that had fallen behind its usage records is corrected. - On Linux, the stats rebuild memory peak that
cctally doctorandcctally db statusreport no longer includes the peak of the process that started the rebuild, which could raise a false 2 GiB memory warning. - With an account selected, a
dashboardConversation Viewer model chip andtranscript search --accountwith--modelno longer fail; they list that account's conversations that used the chosen model family, and the search hits in them. - The
dashboardConversation Viewer's session costs, cost filter and sort, and cache-rebuild counts now update when usage records change or are removed without a transcript change, and after an interrupted sync. cache-sync --prune-orphansno longer fails when a removed session continues in another project, and neither a pricing refresh of the session list norcache-sync --rebuildblocks the conversation store for minutes.- With a Codex budget alert set, Codex commands, the
dashboardand Claude usage recording no longer occasionally stall for about 10 seconds. - The
dashboardfirst-run hint now sits behind any open dialog or sheet, so on a phone it no longer covers the bottom of the Doctor dialog or blocks taps there. - Long same-name project names in the
dashboardProjects panel, its Projects window and a Codex project's detail now shorten the name itself, keeping the(…)part that tells them apart and, on All, the provider prefix visible. - In the
dashboardCodex block and project details,+N morenow shows the extra rows in the dialog itself instead of in a small inner scroll box whose hidden scrollbar made the button look as if it did nothing. - A timed-out
cctally dashboarddiagnosis no longer risks signalling an unrelated process; a new diagnosis starts only after the previous worker and everything it started have exited. - Transcript search no longer misses messages an earlier SQLite build tokenized differently or leaves orphaned entries: the dashboard and
cache-synccheck the search indexes after an upgrade and every 30 days, rebuilding any that disagree. diff --helpand its reference page no longer say--syncrunssync-week; it refreshes the selected providers' cached usage data.- When the
stats.dbmaintenance lock cannot be locked because of a file system error, cctally now names that error instead of saying maintenance is in progress. - When
stats.dbis busy, being upgraded or cannot be read,cctally explainwith a week--windownow reports the store as unavailable (exit 3, or 503 in thedashboard) instead of saying no subscription-week anchor is available. - In the
dashboardProjects panel at window widths of about 1200 to 1440 px, the heading's project count or status no longer runs under thepartialchip; the chip moves to its own line.
v1.120.0
Added
cctally db reindex-searchchecks the conversation search indexes against their messages and rebuilds any that disagree;--checkonly reports.
Changed
- Commands you run now stop waiting on
cache.dborconversations.dbmaintenance after 5 seconds, as they already did forstats.db. Hooks, the status-line writer (record-usage) and background work keep waiting. cctally db statusnow only reads your databases: it no longer resumes an interrupted stats rebuild; the next command that shows or records usage, such asreportor the status line, finishes it instead.- On a large usage history, the
dashboardno longer pauses for most of a second every few seconds while it refreshes, so pages and live updates respond more evenly. It costs about 30 MB more memory. - While
cache.dbis being repaired, thedashboardno longer reads it: conversation search and the reader, Explain and project detail are unavailable until the repair ends, and the Projects panel keeps the data it already shows. - The
dashboardno longer updates or repairs databases while answering a page: a block detail reads stored usage, or your session files during a rebuild, and the dashboard's background work fixes what a page finds. - Right after an upgrade,
db statusanddoctorshow astats.dbfrom an older version as not readable until the next command converts it. - After a gap without a running
cctally dashboard, its transcript catch-up writes about 9% less toconversations.db. Search results are unchanged. - When the Codex hooks that
cctally setupinstalls are enabled, the first Codex turn after deleted Codex sessions are cleaned up no longer redoes that cleanup's quota work, so it stays fast. - With an account selected, the
dashboardConversation Viewer andtranscriptcommands respond much faster: they reuse the conversation summaries they can verify instead of rebuilding the account's whole list. Results are unchanged. - The first cache build after installing, and
cache-sync --rebuild, now write about half as much to disk and finish sooner on large histories. Usage and cost results are unchanged. - Title search in the
dashboardConversation Viewer andtranscript search --kind titleno longer reads every stored message, so it stays fast however large the transcript history is. Results are unchanged.
Fixed
- When
cache.dbexists but cannot be read,cctally pricing-checkno longer says every observed model is priced: it reportsCoverage: unavailable (cache.db could not be read).and a degraded check (cachein--json). - When
cache.dbis found corrupt while an upgrade runs its database steps, cctally no longer deletes it; the next command quarantines a copy and rebuilds it through the normal recovery. - On macOS, remounting the external drive or disk image that holds the data folder no longer makes the next sync rebuild the Codex quota index.
- Codex quota blocks, milestones and alerts stay accurate when updates overlap, Spark windows keep their
dashboardblock detail, and re-enabled alerts notify again. Turning alerts on or changing their rules no longer stalls Codex turns. - After an upgrade, the
dashboardConversation Viewer re-checks its stored session summaries a few at a time, so a session cost that had fallen behind its usage records is corrected. - On Linux, the stats rebuild memory peak that
cctally doctorandcctally db statusreport no longer includes the peak of the process that started the rebuild, which could raise a false 2 GiB memory warning. - With an account selected, a
dashboardConversation Viewer model chip andtranscript search --accountwith--modelno longer fail; they list that account's conversations that used the chosen model family, and the search hits in them. - The
dashboardConversation Viewer's session costs, cost filter and sort, and cache-rebuild counts now update when usage records change or are removed without a transcript change, and after an interrupted sync. cache-sync --prune-orphansno longer fails when a removed session continues in another project, and neither a pricing refresh of the session list norcache-sync --rebuildblocks the conversation store for minutes.- With a Codex budget alert set, Codex commands, the
dashboardand Claude usage recording no longer occasionally stall for about 10 seconds. - The
dashboardfirst-run hint now sits behind any open dialog or sheet, so on a phone it no longer covers the bottom of the Doctor dialog or blocks taps there. - Long same-name project names in the
dashboardProjects panel, its Projects window and a Codex project's detail now shorten the name itself, keeping the(…)part that tells them apart and, on All, the provider prefix visible. - In the
dashboardCodex block and project details,+N morenow shows the extra rows in the dialog itself instead of in a small inner scroll box whose hidden scrollbar made the button look as if it did nothing. - A timed-out
cctally dashboarddiagnosis no longer risks signalling an unrelated process; a new diagnosis starts only after the previous worker and everything it started have exited. - Transcript search no longer misses messages an earlier SQLite build tokenized differently or leaves orphaned entries: the dashboard and
cache-synccheck the search indexes after an upgrade and every 30 days, rebuilding any that disagree. diff --helpand its reference page no longer say--syncrunssync-week; it refreshes the selected providers' cached usage data.- When the
stats.dbmaintenance lock cannot be locked because of a file system error, cctally now names that error instead of saying maintenance is in progress. - When
stats.dbis busy, being upgraded or cannot be read,cctally explainwith a week--windownow reports the store as unavailable (exit 3, or 503 in thedashboard) instead of saying no subscription-week anchor is available. - In the
dashboardProjects panel at window widths of about 1200 to 1440 px, the heading's project count or status no longer runs under thepartialchip; the chip moves to its own line.
v1.118.2
Stable upgrade:
v1.117.2→v1.118.2Includes every released CHANGELOG section in this range, newest first.
[1.118.2] - 2026-11-07
Changed
- The Codex hooks that
cctally setupinstalls no longer re-read the whole quota history to resolve reset anchors, so recording new Codex usage stays fast on a long history.
Fixed
- The dashboard and TUI now regroup a project within about a minute after a git worktree is removed, a repository is created or moved, or a folder becomes a repository; previously they kept the old grouping until restarted.
[1.118.1] - 2026-11-06
Fixed
- With an account selected, the
dashboardConversation Viewer andtranscriptcommands now show conversations' AI titles, and title search ranks, prefix-matches and highlights them; a session spanning two accounts keeps its fallback title. - After a rebuild of the conversation cache, the
dashboardConversation Viewer no longer shows an AI title that a replaced or truncated transcript no longer contains. - The
dashboardConversation Viewer no longer hides a conversation's earlier messages and AI title until the next rebuild after an unfinished last line is trimmed from its transcript.
[1.118.0] - 2026-11-06
Added
- Codex cost reports price
gpt-6.1-terra,gpt-6-rosalindandgpt-6-astra-lawfrom their published rate cards, andgpt-5.5-cyber-previewasgpt-5.5-cyber, instead of falling back togpt-5rates.
Changed
cctally doctorchecks an unchanged journal about three times faster. It keeps a small cache of its journal scan in the journal folder and verifies the journal against it by hash, so every check still sees every byte.- Repairing or fully re-reading Codex history (
cctally cache-sync --source codexor the background worker) no longer locks the cache for the whole run, so Codex hooks keep recording usage during a history repair, and repairs re-read less. - While a Codex history repair runs (
cctally cache-sync --source codexor the background worker), a Claude Code hook that arrives mid-step now waits briefly for its turn, so your Claude usage is recorded right away, not minutes later. - Upgrade rebuilds of
stats.dbandcctally db rederivefinish much faster on large journals, andcctally db rederiveuses less memory. Upgrade rebuilds,db rebuild --db statsanddb rederive --yeswrite far less to disk.
Fixed
- Sessions from a removed git worktree now show and group under their repository in
project, thedashboardand the Conversation Viewer instead of.worktrees; a.gitfolder counts only when it holds a repository. - Codex conversation project labels in the
dashboardConversation Viewer now stay current: a label left on an earlier folder or(unassigned)is corrected on the next full refresh. - Anonymized Codex
transcript exportoutput now scrubs both the repository name and the session folder's own name, as Claude exports already do. - The
dashboardno longer rescans every Claude transcript on every refresh when its cache still tracks transcripts that were deleted from disk. - Compressed Codex history that fails conversation verification no longer keeps the
dashboardrescanning every Claude and Codex transcript on each refresh; that history keeps its last verified view until it changes. - Upgrading from a cctally older than 1.74.0 on a Python without a zstd decoder keeps cached Codex history in
codex-*reports; it is fully refreshed once Python 3.14+ orbackports.zstdis available. - The dashboard's Doctor modal and
cctally doctornow wait at most half a second in total for stats.db maintenance, instead of about 5 seconds per database read (40 to 60 seconds per report), and report the maintenance. cctally doctor's Latest snapshot check no longer fails with "Hooks may be broken" when usage stays flat for over an hour; it now counts every accepted reading, and says when stats.db cannot be read.- The Current Week freshness chip in
cctally dashboardandcctally tuino longer turns stale while usage stays flat; it now ages from the newest reading that confirms the shown percentage. - An interrupted
cctally cache-sync --source codex --rebuildno longer gives Codex usage from before account tracking to the account logged in at the retry; it stays unattributed. A rebuild with an unreadable rollout now stays pending. - A running
dashboardnow shows corrected usage, milestone and block figures after a stored correction (db rederive, a rebuild or an in-place fix) without a new sample, and a reopened Claude or Codex cycle detail shows current figures. explain's short-conversations-with-large-context class no longer measures prior-generation Claude requests above 200K tokens against a 200K window, and reports a request it cannot place as undecided instead of guessing.cctally dashboard --no-syncstarted while stats.db is being rebuilt after an upgrade or repaired after corruption now shows your data once that finishes, instead of staying empty until a restart.- Conversation Viewer search results in the
dashboardnow keep a transcript's own square brackets, such as[WARN], as text and highlight only the words your search matched. - The
dashboardConversation Viewer header now shows a Codex conversation's project and git branch, orno branch recordedwhen Codex saved none, instead of—, as it does for a Claude conversation opened from All. - The
statuslineno longer runs at full CPU for minutes after recording a reading on macOS, and Claude Code hooks no longer pile up behind it: ahook-tickthat finds another usage refresh already in progress skips its own. - Removing cached Codex history for deleted rollout files (
cctally cache-syncand the background cleanup) and recording Codex conversation updates no longer slow down as history grows; a 20-minute cleanup now takes a few minutes.
v1.118.1
Fixed
- With an account selected, the
dashboardConversation Viewer andtranscriptcommands now show conversations' AI titles, and title search ranks, prefix-matches and highlights them; a session spanning two accounts keeps its fallback title. - After a rebuild of the conversation cache, the
dashboardConversation Viewer no longer shows an AI title that a replaced or truncated transcript no longer contains. - The
dashboardConversation Viewer no longer hides a conversation's earlier messages and AI title until the next rebuild after an unfinished last line is trimmed from its transcript.
v1.118.0
Added
- Codex cost reports price
gpt-6.1-terra,gpt-6-rosalindandgpt-6-astra-lawfrom their published rate cards, andgpt-5.5-cyber-previewasgpt-5.5-cyber, instead of falling back togpt-5rates.
Changed
cctally doctorchecks an unchanged journal about three times faster. It keeps a small cache of its journal scan in the journal folder and verifies the journal against it by hash, so every check still sees every byte.- Repairing or fully re-reading Codex history (
cctally cache-sync --source codexor the background worker) no longer locks the cache for the whole run, so Codex hooks keep recording usage during a history repair, and repairs re-read less. - While a Codex history repair runs (
cctally cache-sync --source codexor the background worker), a Claude Code hook that arrives mid-step now waits briefly for its turn, so your Claude usage is recorded right away, not minutes later. - Upgrade rebuilds of
stats.dbandcctally db rederivefinish much faster on large journals, andcctally db rederiveuses less memory. Upgrade rebuilds,db rebuild --db statsanddb rederive --yeswrite far less to disk.
Fixed
- Sessions from a removed git worktree now show and group under their repository in
project, thedashboardand the Conversation Viewer instead of.worktrees; a.gitfolder counts only when it holds a repository. - Codex conversation project labels in the
dashboardConversation Viewer now stay current: a label left on an earlier folder or(unassigned)is corrected on the next full refresh. - Anonymized Codex
transcript exportoutput now scrubs both the repository name and the session folder's own name, as Claude exports already do. - The
dashboardno longer rescans every Claude transcript on every refresh when its cache still tracks transcripts that were deleted from disk. - Compressed Codex history that fails conversation verification no longer keeps the
dashboardrescanning every Claude and Codex transcript on each refresh; that history keeps its last verified view until it changes. - Upgrading from a cctally older than 1.74.0 on a Python without a zstd decoder keeps cached Codex history in
codex-*reports; it is fully refreshed once Python 3.14+ orbackports.zstdis available. - The dashboard's Doctor modal and
cctally doctornow wait at most half a second in total for stats.db maintenance, instead of about 5 seconds per database read (40 to 60 seconds per report), and report the maintenance. cctally doctor's Latest snapshot check no longer fails with "Hooks may be broken" when usage stays flat for over an hour; it now counts every accepted reading, and says when stats.db cannot be read.- The Current Week freshness chip in
cctally dashboardandcctally tuino longer turns stale while usage stays flat; it now ages from the newest reading that confirms the shown percentage. - An interrupted
cctally cache-sync --source codex --rebuildno longer gives Codex usage from before account tracking to the account logged in at the retry; it stays unattributed. A rebuild with an unreadable rollout now stays pending. - A running
dashboardnow shows corrected usage, milestone and block figures after a stored correction (db rederive, a rebuild or an in-place fix) without a new sample, and a reopened Claude or Codex cycle detail shows current figures. explain's short-conversations-with-large-context class no longer measures prior-generation Claude requests above 200K tokens against a 200K window, and reports a request it cannot place as undecided instead of guessing.cctally dashboard --no-syncstarted while stats.db is being rebuilt after an upgrade or repaired after corruption now shows your data once that finishes, instead of staying empty until a restart.- Conversation Viewer search results in the
dashboardnow keep a transcript's own square brackets, such as[WARN], as text and highlight only the words your search matched. - The
dashboardConversation Viewer header now shows a Codex conversation's project and git branch, orno branch recordedwhen Codex saved none, instead of—, as it does for a Claude conversation opened from All. - The
statuslineno longer runs at full CPU for minutes after recording a reading on macOS, and Claude Code hooks no longer pile up behind it: ahook-tickthat finds another usage refresh already in progress skips its own. - Removing cached Codex history for deleted rollout files (
cctally cache-syncand the background cleanup) and recording Codex conversation updates no longer slow down as history grows; a 20-minute cleanup now takes a few minutes.
v1.117.2
Stable upgrade:
v1.116.1→v1.117.2Includes every released CHANGELOG section in this range, newest first.
[1.117.2] - 2026-11-03
Fixed
- With more than one Claude account, the
cctally dashboardCurrent Week card and modal show one account's cycle, milestones, spend and$/1%beside that account's label, and the older/newer cycle buttons stay on that account. cctally five-hour-breakdownand the dashboard's five-hour milestones show—(nullin--json) instead of a weekly percentage that a weekly credit later retired.- With more than one Claude account, the
cctally dashboardAll view shows its combined spend headline, summing each account's current cycle, instead of usually leaving it out. - Under Python 3.11,
cctally quota --jsonno longer reports family shares and calibration figures that differ in their last digits from newer Python versions.
[1.117.0] - 2026-11-02
Added
cctally doctoraddsaccounts.claude_cost_attribution, reporting how recent Claude cost was attributed to accounts and the progress of the one-time upgrade step that records existing attribution.
Changed
cctally quotano longer withholds a forecast over a small Sonnet 5.5 share shift that meter evidence shows to be immaterial, and withheld lines name the family reason. Stored calibrations go stale until the nextcctally quotarun.- When a reviewed
cctally db rederivecorrection changes a calibration's inputs,cctally forecastand the dashboard withhold calibrated figures ("recalibration needed") untilcctally quotarefits. Stats rebuilds once on upgrade. - A metering-rate change fitted from corrected inputs no longer shows its status-line marker or sends its notification; the dashboard's Recent Alerts keeps it, labelled "superseded by a correction".
- After a gap without a running
cctally dashboard, its first transcript catch-up writes less toconversations.db, andcctally dashboard-perfandcctally doctorno longer judge that catch-up as steady-state disk writes. - Automatic credit, early-reset and status-line drop detection paused by an older cctally process now resumes on its own, per account and per limit;
cctally doctorshows what each limit waits for. Stats rebuilds once on upgrade.
Fixed
- A
stats.dbupgrade rebuild orcctally db rebuild --db statsno longer reprices closed five-hour blocks and their milestones at the current rate card when it recovers missing records. - Archived Codex threads now count in
codex-*reports, quota history and the dashboard, and archiving no longer deletes history. Archived usage first seen with no recorded account stays unattributed unless quota windows assign it. - On Python builds that link SQLite 3.54.1, a running
cctally dashboardreclaims transcript space freed by retention again, soconversations.dbshrinks instead of keeping the freed pages. - An idle Claude Code session's stale
cctally statuslinereading no longer records a false weekly reset, five-hour credit or zero-cost milestone once a later usage check reads higher; the stats index rebuilds once after upgrading. - Claude cost is attributed to the account active when Claude Code wrote it, not the one active when cctally read it. Upgrades and
cctally cache-sync --rebuildkeep each row's account. cctally statuslinemeasures🧠context % against the session's real context window (1M for Opus, Sonnet and Haiku 5.5), so an Opus 5.5 session no longer shows a red percentage five times too high and Fable sessions no longer showN/A.- A
stats.dbupgrade rebuild could exit early as busy when a status-line check for a running rebuild landed at the instant it started, delaying the rebuild; that check no longer interferes with a starting rebuild. cctally explainand the dashboard's Explain view measure Opus, Sonnet and Haiku 5.5 requests against their 1M context window and now evaluate Fable requests, so short conversations are no longer flagged for large context against 200K.cctally five-hour-blocks,cctally five-hour-breakdownand the dashboard keep each five-hour window's genuine peak, milestones and readings through a usage credit and through every stats.db rebuild, instead of dropping some.cctally record-creditrecords the exact--attime instead of the start of that hour, keeping readings taken earlier in that hour, and a--forcereplacement at the replaced credit's own second now takes effect.- A weekly usage credit is no longer lost when recording is interrupted while it is being saved.
- After stats.db is rebuilt from a missing or corrupt file, the stats index keeps updating incrementally instead of falling back to rereading the whole recorded history.
- On an exFAT data folder, cctally now quarantines and rebuilds a corrupt
conversations.dborcache.dband can rundb repair --db stats, and a rebuild no longer fails after 20 s while another cctally process has the store open but idle. - On macOS,
cctally cache-sync --rebuildno longer writes a full copy ofconversations.dbon volumes without copy-on-write clones (HFS+, exFAT); there and on Linux ext4 its integrity check runs in place when no write-ahead log is pending. cctally dashboardno longer waits on a reverse-DNS lookup of its bind address before it prints its URL and serves, a wait that lasted the resolver's whole timeout where the network never answers that lookup.- Codex hooks installed by
cctally setupno longer stall Codex sessions for over 10 seconds on large histories or after upgrading, or for over a minute with Codex budget alerts on; leftover work runs in the background. cctally db rederiverestores a five-hour milestone's cost when an earlier rederive repriced it under a closed block; one whose history is ambiguous keeps its cost and is listed inwithheldMilestoneRepairs.
v1.117.0
Added
cctally doctoraddsaccounts.claude_cost_attribution, reporting how recent Claude cost was attributed to accounts and the progress of the one-time upgrade step that records existing attribution.
Changed
cctally quotano longer withholds a forecast over a small Sonnet 5.5 share shift that meter evidence shows to be immaterial, and withheld lines name the family reason. Stored calibrations go stale until the nextcctally quotarun.- When a reviewed
cctally db rederivecorrection changes a calibration's inputs,cctally forecastand the dashboard withhold calibrated figures ("recalibration needed") untilcctally quotarefits. Stats rebuilds once on upgrade. - A metering-rate change fitted from corrected inputs no longer shows its status-line marker or sends its notification; the dashboard's Recent Alerts keeps it, labelled "superseded by a correction".
- After a gap without a running
cctally dashboard, its first transcript catch-up writes less toconversations.db, andcctally dashboard-perfandcctally doctorno longer judge that catch-up as steady-state disk writes. - Automatic credit, early-reset and status-line drop detection paused by an older cctally process now resumes on its own, per account and per limit;
cctally doctorshows what each limit waits for. Stats rebuilds once on upgrade.
Fixed
- A
stats.dbupgrade rebuild orcctally db rebuild --db statsno longer reprices closed five-hour blocks and their milestones at the current rate card when it recovers missing records. - Archived Codex threads now count in
codex-*reports, quota history and the dashboard, and archiving no longer deletes history. Archived usage first seen with no recorded account stays unattributed unless quota windows assign it. - On Python builds that link SQLite 3.54.1, a running
cctally dashboardreclaims transcript space freed by retention again, soconversations.dbshrinks instead of keeping the freed pages. - An idle Claude Code session's stale
cctally statuslinereading no longer records a false weekly reset, five-hour credit or zero-cost milestone once a later usage check reads higher; the stats index rebuilds once after upgrading. - Claude cost is attributed to the account active when Claude Code wrote it, not the one active when cctally read it. Upgrades and
cctally cache-sync --rebuildkeep each row's account. cctally statuslinemeasures🧠context % against the session's real context window (1M for Opus, Sonnet and Haiku 5.5), so an Opus 5.5 session no longer shows a red percentage five times too high and Fable sessions no longer showN/A.- A
stats.dbupgrade rebuild could exit early as busy when a status-line check for a running rebuild landed at the instant it started, delaying the rebuild; that check no longer interferes with a starting rebuild. cctally explainand the dashboard's Explain view measure Opus, Sonnet and Haiku 5.5 requests against their 1M context window and now evaluate Fable requests, so short conversations are no longer flagged for large context against 200K.cctally five-hour-blocks,cctally five-hour-breakdownand the dashboard keep each five-hour window's genuine peak, milestones and readings through a usage credit and through every stats.db rebuild, instead of dropping some.cctally record-creditrecords the exact--attime instead of the start of that hour, keeping readings taken earlier in that hour, and a--forcereplacement at the replaced credit's own second now takes effect.- A weekly usage credit is no longer lost when recording is interrupted while it is being saved.
- After stats.db is rebuilt from a missing or corrupt file, the stats index keeps updating incrementally instead of falling back to rereading the whole recorded history.
- On an exFAT data folder, cctally now quarantines and rebuilds a corrupt
conversations.dborcache.dband can rundb repair --db stats, and a rebuild no longer fails after 20 s while another cctally process has the store open but idle. - On macOS,
cctally cache-sync --rebuildno longer writes a full copy ofconversations.dbon volumes without copy-on-write clones (HFS+, exFAT); there and on Linux ext4 its integrity check runs in place when no write-ahead log is pending. cctally dashboardno longer waits on a reverse-DNS lookup of its bind address before it prints its URL and serves, a wait that lasted the resolver's whole timeout where the network never answers that lookup.- Codex hooks installed by
cctally setupno longer stall Codex sessions for over 10 seconds on large histories or after upgrading, or for over a minute with Codex budget alerts on; leftover work runs in the background. cctally db rederiverestores a five-hour milestone's cost when an earlier rederive repriced it under a closed block; one whose history is ambiguous keeps its cost and is listed inwithheldMilestoneRepairs.
v1.116.1
Stable upgrade:
v1.113.0→v1.116.1Includes every released CHANGELOG section in this range, newest first.
[1.116.1] - 2026-10-28
Maintenance
- Internal improvements only; no user-facing changes.
[1.116.0] - 2026-10-28
Added
cctally doctor,cctally db statusand the dashboard's Doctor chip now say when a stats.db upgrade rebuild or auto-heal is running, has stalled, failed or died, or used more memory than its bound.
Changed
- The stats.db upgrade rebuild that runs after an update,
cctally db rebuildandcctally db rederiveuse far less memory and disk, and the upgrade rebuild finishes faster.
Fixed
- After a usage credit,
cctally statuslinerefuses an idle session's cached usage once it is proven stale, and five-hour views stop hiding genuine weekly values.cctally db rederiveaddsretract; stats.db rebuilds once after upgrading. - Under Python 3.11,
cctally quota --jsonno longer reports a mix-support radius that differs in its last digits from one run to the next. cctally quota --jsonreports the sameeligibilityFenceon Linux as on macOS; on Linux it could differ in its last digits.- A sync with nothing new to read no longer rewrites the compressed Codex rollout status in
cache.dbandconversations.db, so an idlecctally dashboardwrites less to disk. cctally cache-sync --rebuildrebuilds transcripts again on a filesystem without copy-on-write clones, such as ext4 on Linux, instead of failing every time.- On Python 3.11, the upgrade that moves transcripts out of
cache.dbnow shrinks the file as intended. If you already upgraded on Python 3.11, runcctally db vacuum --db cacheto reclaim the space.
[1.115.0] - 2026-10-21
Added
cctally doctorhas two new checks:Legacy writerwarns while an older cctally process left running across the upgrade is still writing or its status-line files cannot be read, andOutcome quarantinenames each paused account and why.cctally quotarecognises Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 and weights usage from 2026-09-22 with Opus 5.5 coefficients, fitting it when the model-mix checks pass; a week spanning that date is not modelled.
Changed
- The status line now records every reading under the Claude account active when it was captured, so switching accounts can no longer file one account's usage under another.
- Upgrading marks a one-time cutover: anything an older cctally process writes afterwards, including after a downgrade and re-upgrade, is kept out of your usage history. Restart such processes, usually a
cctally dashboard. - After an older process writes post-upgrade, credit and reset detection pauses until each account reads clean, or indefinitely if status-line files changed or are unreadable; usage still records, and
cctally record-creditrecords resets. cctally record-usageexits 2 and records nothing when the active Claude account cannot be read consistently, instead of recording under a guessed account.- Existing
cctally quotacalibrations become incompatible immediately; each account's is replaced on its next successful automatic fit, and a diagnostic run or too little Opus 5.5 history leaves the old one stored and ignored.
Fixed
- Codex reports now read compressed rollouts (
.jsonl.zst) without losing cached history, slow history cleanup no longer runs inside the Codex hook, andcctally doctorhas a newCompressed Codex rolloutscheck. daily,sessionand other Claude cost reports price Sonnet 5.5 cache reads at $0.10 per million tokens instead of $0.20. Recorded snapshots, closed blocks and milestones keep their original amounts.- The status line's 5-hour figure no longer changes, or retries a pending drop, when another account's 5-hour reset is recorded.
cctally db rebuild --db statsno longer drops percent milestones after a 5-hour or manual credit, which also stalled recording afterwards; this upgrade's one-time rebuild restores them.- A usage reading recorded right after an upgrade while an older cctally process (such as a dashboard) was still running, or during
cctally db rebuild --db statsor a stats index repair, no longer goes missing or waits for a later reading. cctally db rebuild --db statskeeps the current 5-hour block's first-seen time and pricing details as recorded. It no longer moves that time after a 5-hour credit or when two readings were recorded out of order.- The status line, dashboard, TUI,
forecast,five-hour-blocksandfive-hour-breakdownshow the latest 5-hour reading when the newest reading has none, instead of a blank or 0 % 5-hour figure or a closed current block. - A 5-hour credit no longer discards a weekly reading recorded alongside the stale 5-hour figure, so the weekly figure no longer falls back to an older value.
- A weekly credit detected mid-week no longer discards the 5-hour reading recorded with the stale weekly figure, so the status line, 5-hour blocks and 5-hour milestones keep it, and upgrading restores readings already lost.
[1.114.0] - 2026-10-08
Added
cctally dashboard-perfreports the dashboard's disk writes per minute and per publication against a write budget, andcctally doctorwarns when a running dashboard or transcript maintenance exceeds it.- Claude cost reports price
claude-haiku-5-5at its published rates, including the higher rates for a request whose prompt exceeds 100,000 tokens, instead of showing $0.
Changed
cctally db vacuum --db cacheor--db conversationsstarted while the dashboard runs now asks it to release the store and waits up to 20 seconds before vacuuming, or refusing with the same message as before.cctally doctorno longer warns about modest free space inconversations.db; it warns only while a large backlog (over 2 GiB and 20% of the file) is being reclaimed at a paced rate, and says why when reclaim is refused.
Fixed
- Statusline and usage recording avoid redundant sorting of retained snapshots and expired blocks, with the same quota readings and publication order.
- The
dashboardblock cost chart shows every trajectory sample when it exceeds a retained total. - The
dashboardSessions filter and sort icons keep their intended size on phones. - The
dashboardkeeps same-name project suffixes and panel counts readable, contains account status chips on phones, and keeps the swipe cue clear of spend details. Returning to All accounts starts at the first card. cctally db rederivekeeps five-hour milestone marginal costs consistent when applying reviewed weekly decisions, and resolves held-close corrections in one apply so the next preview needs no further correction on identical inputs.cctally dashboardCodex panels now match a fresh rebuild after a memory-cache reset, a brief metadata or account-registry read error, a past-window share, or a window move under a Codex budget, instead of doubling, going stale or failing.cctally db rederivenow reports acache.dbthat is missing a column it needs asmissing-source(exit 2), naming the column, instead of failing with an internal error (exit 3).cctally db rederiveno longer fails with an internal error whencache.dbstill holds a transcript search index that this Python's SQLite cannot load.- Milestone tables put a negative amount's sign before the dollar sign:
−$71.21in the dashboard cycle modal,-$71.21in the TUI, and-$71.210000infive-hour-breakdown,percent-breakdownandreport --detail, not$-71.21. cctally dashboardno longer writes about a gigabyte a minute to disk on a large history: steady-state writes now follow new transcript data, and removing expired transcripts and reclaiming their space are paced.cctally dashboard,cctally cache-syncand the Codex hook no longer write SQLite temp files while syncing transcripts and quota data, and two background lookups no longer re-sort a whole session's or quota window's history.
v1.116.0
Added
cctally doctor,cctally db statusand the dashboard's Doctor chip now say when a stats.db upgrade rebuild or auto-heal is running, has stalled, failed or died, or used more memory than its bound.
Changed
- The stats.db upgrade rebuild that runs after an update,
cctally db rebuildandcctally db rederiveuse far less memory and disk, and the upgrade rebuild finishes faster.
Fixed
- After a usage credit,
cctally statuslinerefuses an idle session's cached usage once it is proven stale, and five-hour views stop hiding genuine weekly values.cctally db rederiveaddsretract; stats.db rebuilds once after upgrading. - Under Python 3.11,
cctally quota --jsonno longer reports a mix-support radius that differs in its last digits from one run to the next. cctally quota --jsonreports the sameeligibilityFenceon Linux as on macOS; on Linux it could differ in its last digits.- A sync with nothing new to read no longer rewrites the compressed Codex rollout status in
cache.dbandconversations.db, so an idlecctally dashboardwrites less to disk. cctally cache-sync --rebuildrebuilds transcripts again on a filesystem without copy-on-write clones, such as ext4 on Linux, instead of failing every time.- On Python 3.11, the upgrade that moves transcripts out of
cache.dbnow shrinks the file as intended. If you already upgraded on Python 3.11, runcctally db vacuum --db cacheto reclaim the space.