中文
← Back to home

Release notes

Every change to Aiming Cookie, in one place.

v1.4.18

Latest2026-10-20
  • FixedFixed an overly strict telemetry check that mis-rejected fast-fire scenarios (e.g. 1w1ts): analysis no longer falls back to the slower path, and full-depth telemetry evidence is restored
  • FixedFixed sign-in occasionally hanging after authorization: the app now polls for the login result as a fallback when the confirmation event gets lost in the OS handoff

v1.4.17

2026-10-19
  • FixedFixed xAI (Grok) sign-in failures: the app now connects to the login service through your system proxy, so device-code authorization works on networks that require one
  • FixedDetailed reasons for sign-in failures are now written to the diagnostic log and included in support bundles, making issues easier to pin down
  • FixedCompleted the 1.4.15 encoding fix: reading collector output during training finalization could still crash on non-Chinese Windows — now fixed as well
  • FixedFixed a per-run telemetry freeze bug that dropped flag rows and skewed target life-boundary statistics
  • FixedFixed a rare race where a training record's state could be rolled back by a stale snapshot during concurrent finalization writes
  • FixedHonest stop states: while a turn is still winding down the UI shows a stopping state instead of falsely marking it stopped
  • FixedQueued messages no longer leak across sessions — content queued in chat A can no longer be sent to chat B
  • FixedAn unfinished storage migration now blocks changing the location again, with clear recovery guidance (free the locked files and restart to resume)
  • FixedProvenance notes for telemetry-derived traces (source and skipped rows) now reach every metric and the overview instead of staying top-level only

v1.4.15

2026-10-13
  • FixedBYOK Anthropic users: newer models like claude-haiku-5.5 no longer fail with a protocol error — the app adapts and retries automatically
  • FixedFixed telemetry collector crashes on non-Chinese Windows systems (root cause of dead external telemetry for some international users)
  • FixedNew guardrail against corrupted model output: automatic retry with a clear notice, plus a regenerate entry on affected replies
  • FixedInput containing corrupted characters (likely an IME glitch) is caught before sending and returned for retyping

v1.4.14

2026-10-12
  • FixedNo more black flash on startup: the window now appears only after the first frame is painted

v1.4.13

2026-10-12
  • FixedReplies cut off by an app restart are now reported honestly with one-click retry: no more misleading "reply stopped" label — interrupted replies get an "ask again" button
  • FixedInstalling updates or changing the storage location now checks whether Coach is mid-reply: while a reply is generating, it asks you to wait so a restart won't cut it off

v1.4.12

2026-10-12
  • NewRedesigned first-run activation: if your trial credit is still being provisioned on first use, an initialization indicator shows and the app resumes automatically instead of wrongly asking you to pay
  • ImprovedTraining history rows now show a diagnosis-type badge and a one-line takeaway, so each run's focus and conclusion are visible at a glance
  • NewAnalysis artifacts now support an auto-cleanup cap (default 10GB, adjustable or off): only analysis artifacts are cleaned — recordings and raw traces are never touched automatically, and every auto-clean is logged in Settings
  • FixedCoach session files now self-repair when damaged: history conversations no longer get lost (corrupt content is quarantined as evidence, the session opens normally)
  • ImprovedFaster analysis: redundant re-verification of the same video removed and hot loops vectorized; the capture/recording hot path uses fewer resources
  • FixedClosing the window responds instantly: data is finalized safely in the background with timeout guards — no more long unresponsive waits after clicking close
  • FixedThe onboarding success banner no longer renders in the red error style
  • FixedThe startup window now has the same rounded corners as the main window
  • FixedMore resilient startup: no more crashes when antivirus interferes; internal waits now have timeout guards so nothing hangs forever
  • FixedProvider authorization polling no longer stalls after a single network hiccup; the capture input channel self-heals and reports honestly

v1.4.11

2026-10-09
  • FixedFixed crosshair/target trajectory data being wrongly discarded on static-respawn clicking maps (the 1wall5targets / beanClick family): trajectories on these maps are now rebuilt correctly and feed the analysis
  • ImprovedMore data self-healing: raw mouse data wrongly dropped by earlier versions keeps recovering from local files automatically, and stale 'available' records left by storage migration are now corrected and honestly labeled

v1.4.10

2026-10-09
  • FixedFixed raw mouse data showing as unavailable for recent runs after an update or restart, which degraded analyses to scores and video only; affected past runs recover automatically after updating - no need to replay them

v1.4.9

2026-10-08
  • FixedFixed training analysis hanging forever or failing on some PCs - the full analysis now completes
  • NewBroader analysis compatibility: full analyses that reconcile exactly with official KovaaK stats, no longer depending on KovaaK's optional performance record file

v1.4.8

2026-10-08
  • NewThe coach now reconnects over a backup route automatically when the network is uncooperative - no setup needed on your side
  • FixedFixed network failures being misreported as model errors: the app now clearly points to the network and offers one-tap retry

v1.4.7

2026-10-08
  • FixedFixed coach conversations failing to save on some machines: when the log file is briefly locked by cloud sync or antivirus, we now wait and retry instead of interrupting the chat

v1.4.6

2026-10-08
  • NewThe coach now reads each scenario first - what it trains, how it scores, how targets spread - so advice fits the map: pacing advice on speed scenarios, settle-and-confirm only on precision ones
  • NewSmarter practice picks: classic standard maps you already have installed come first, then subscription suggestions, or a targeted pick for your weak spot
  • NewAnalyses now include a short scenario profile: the skill it trains, its scoring rules, and target layout at a glance
  • FixedKnowledge base revision: a few cited scenario values corrected - what the coach cites is now the verified version

v1.4.5

2026-10-07
  • NewFixed missing mouse-trace and target data on some machines: capture keeps working across game restarts
  • FixedRe-processing history after an app restart no longer runs piles of useless work — results show up faster
  • NewHit-detection radii are now backfilled from real official scenario data, making geometry metrics more accurate
  • ImprovedSessions outside the capture retention window are now honestly labeled in their details

v1.4.4

2026-10-07
  • NewMetrics for multi-target strafing scenarios (e.g. Cata IC Long Strafes) are fixed and back in a trustworthy range
  • FixedCoach reads results honestly: says clearly when a session has no flagged issues, and no longer invents scenario names or ratings
  • FixedRemoved misleading wording that described normal data as degraded or limited
  • ImprovedScenario recommendations no longer repeat dev-template scenarios

v1.4.3

2026-10-07
  • NewResizing the game window or switching resolution mid-session no longer interrupts recording — your training data is kept
  • FixedAnalysis results for sessions where the window was resized no longer drift due to letterbox bars
  • FixedWhen game data can't be read, the app now explains why (e.g. KovaaK running as administrator, antivirus interference) and what to do
  • ImprovedCoach errors are now clearer: quota exhausted, network hiccups and busy service are explained separately, with a one-click top-up when quota runs out
  • ImprovedVery long conversations are compacted automatically to keep reply quality, with a visible progress note while it happens
  • FixedRemoved inaccurate auto-renewal wording from the member page (purchases are one-time per month)

v1.4.2

2026-10-06
  • NewStartup is now visible: a launch screen with logo and progress appears first, slow starts explain why, and failures come with clear guidance
  • FixedFixed the app freezing and crashing on startup failure: it now stays open with a one-click diagnostics export
  • FixedFixed sessions failing outright when camera frames were missing during processing

v1.4.1

2026-10-06
  • NewSession motion-data processing now carries forward: re-processing a session keeps its previously successful result instead of reworking it from scratch
  • FixedFixed sessions flipping from "processed" back to "failed" when processed again
  • ImprovedA few borderline sessions that were previously rejected outright now process successfully

v1.4.0

2026-10-05
  • NewMotion analysis for target-switching scenarios: switch transition time (from one kill to re-acquiring the next target) and switch path efficiency, measured from ground-truth telemetry rather than estimation
  • FixedFixed analyses getting stuck on "queued" and never finishing: static-clicking scenarios (e.g. 1wall 6targets small) now complete on the first try, and previously stuck analyses recover on retry
  • ImprovedSmoother first-run onboarding: the coach gets straight to training instead of circling
  • FixedThree membership-page fixes: plan-card colors in dark/light themes, continuous-billing wording now matches the actual monthly purchase, and the account card no longer renders misaligned
  • FixedMotion-analysis time alignment fixed: all motion metrics now align to the real round window — previously the window started about 8 seconds early, cutting off the end of each round and mixing in pre-round data
  • FixedTarget track segmentation fixed: kill counts now match the game's official stats line by line — previously targets were over-segmented, inflating kills by nearly half and distorting tracking metrics
  • ImprovedData-quality transparency: analyses with imperfect data now flag which metrics deserve caution, and the coach says so plainly instead of presenting compromised numbers as reliable readings
  • ImprovedScenario classification fix: switching-style scenarios (such as Tile Frenzy 180) are now analyzed as switching, with switch cadence and path readings of their own instead of tracking-style yardsticks

v1.3.9

2026-10-05
  • NewTraining-record processing is now visible at a glance: rows show what is still being processed and how many runs are queued, and a notification pops up when everything is ready
  • FixedTracking-style motion analysis is dramatically faster and no longer fails intermittently: no more long waits ending in a score-only fallback - crosshair-to-target adherence and error metrics now arrive within seconds, and repeated analysis of the same run gives consistent results
  • FixedFixed some new users getting stuck on sign-in onboarding: the one-time credential from the browser sign-in is no longer lost — it is retried automatically if the app isn't ready, and reopening the app picks the earlier sign-up back up
  • FixedMore resilient startup: page-load failures now retry instead of leaving a rare blank screen, and launching a second copy no longer blanks the first window
  • FixedMotion-layer diagnostics fixed: tracking-style scenarios now produce crosshair-versus-target motion analysis instead of being skipped, and one faulty round no longer drags down the rest of a session
  • FixedFixed “larger + slowed” style scenarios failing to process their data: targets carried over from the previous run are no longer mistaken for on-screen debris, so these scenarios process and analyze normally again

v1.3.8

2026-10-04
  • FixedNew-user funnel fixes: the official Aiming Cookie option now leads the provider list; the client polls automatically after sign-in instead of waiting forever; the trial CTA appears on its own once quota lands — no restart needed
  • FixedDevice handoff validity extended from 10 to 30 minutes, leaving ample time for email verification

v1.3.7

2026-10-03
  • FixedFixed new trial users being unable to finish onboarding after choosing the free trial — the enter-workspace button no longer stays disabled

v1.3.6

2026-10-03
  • FixedSecurity hardening: the local model-service API now requires a launch-scoped access token, blocking cross-origin credential reads from web pages
  • FixedFixed corrupted session files breaking the training history list
  • FixedFixed a rare startup white-screen: corrupted local cache is now detected and discarded automatically
  • FixedFixed a rare case where changing the storage location could make data appear missing
  • FixedFixed a crash when analyzing runs with sparse telemetry frames
  • ImprovedVideo verification during analysis now runs off the main thread for a smoother UI
  • ImprovedNew KovaaK installation guide card, plus interface motion and interaction polish

v1.3.5

2026-10-02
  • FixedFixed training video recording on certain Windows Insider preview builds
  • FixedImproved recording compatibility on virtual displays and remote desktop setups
  • FixedFixed very long conversations failing or burning excess quota when first resumed: history is now compacted automatically before sending
  • ImprovedSelf-healing restarts of the capture engine are now recorded in diagnostics for faster remote troubleshooting
  • FixedFixed new users getting stuck on the model-service step: you can now start with the free trial first and subscribe after completing one verification analysis

v1.3.4

2026-10-01
  • FixedCoach no longer returns an occasional empty reply: it now retries automatically once, so you don't have to ask again by hand
  • FixedIf the selected model becomes unavailable, the app now repairs it automatically and points you to what to do next instead of getting stuck on the error
  • ImprovedError messages for gateway quota issues are now clearer, so you can tell at a glance what went wrong

v1.3.3

2026-09-30
  • NewScreen recording now supports non-standard window sizes: odd pixel dimensions from display scaling or virtual displays are handled automatically instead of failing
  • FixedTelemetry capture subprocesses no longer fail silently: crash reasons are recorded in diagnostics, and transient failures recover automatically
  • ImprovedCapture is more resilient in adverse conditions such as game-exit races or removable-drive dropouts, without busy-looping
  • ImprovedDiagnostics bundles now include capture window size, encode size, per-layer encoder failure details and capture subprocess crash logs for faster remote troubleshooting

v1.3.2

2026-09-30
  • NewTraining history now supports deleting a record: hover a row and click the trash icon twice to remove it
  • FixedDeep analysis is now available right after each run, without quitting KovaaK first
  • FixedFixed screen capture failing to start on some multi-GPU setups; capture now retries automatically on another GPU
  • FixedFixed conversations failing when using your own DeepSeek API key on the official endpoint

v1.3.1

2026-09-29
  • FixedCompatibility fix for some older AMD GPUs — recording now falls back to software encoding instead of failing entirely
  • FixedFixed all later runs failing to analyze after resizing the game window mid-session
  • FixedFixed run tracking going silent when the app starts while Steam is updating KovaaK

v1.3.0

2026-09-28
  • NewCustom data location — move recordings and telemetry off your system drive
  • NewUsage records in Account Center — time, model and tokens for every AI call
  • ImprovedLong-conversation memory fix — full history retained with cache-backed replies
  • FixedFixed rapid-fire messages in a new chat being split across sessions
  • FixedFixed connection test falsely reporting expired credentials

v1.2.2

2026-09-18
  • FixedTop-bar discussion chips no longer overlap the three window buttons
  • ImprovedMore stable install and update flow

v1.2.1

2026-09-14
  • NewOfficial scenario library search (fuzzy match across 175k scenarios)