Settings > Interface > PRESETS lists extra preset folders with an ADD FOLDER… picker. Every listed folder is scanned as user presets, for a shared drive or a cloud library.
A migrator for KURV 0.8 `.kurv` presets: `cargo run --example migrate_kurv_presets -- <in.kurv>... --out <dir>` writes 0.20 `.kurvy` files and reports what carried over. In-group RATIO BRICKWALL filters move into the WARPs rack; the retired automation banks are dropped.
Fixed
Loading a preset restores every control. The browser wrote the file's host-automation bank over the patch and skipped the restore hook, so oscillator levels, filter cutoffs and group gains bound to an automation slot landed at minimum.
Factory presets show `factory` in the SAVED column instead of an epoch date.
Windows VST3 loads in hosts that scan without searching the plugin's own folder, FL Studio among them. The MinGW runtime (libstdc++, libgcc, winpthread) is linked statically, so no DLLs ship beside KURV.vst3.
PM/FM now works on an oscillator that carries a prepared WARP (SPECTRAL, DISPERSE, TILT, RANDOM) or SYNC. A prepared spectrum is read by phase, so the per-sample phase offset lands at the table read; the refusal "Phase modulation cannot be combined with this WARP" is gone from the routing matrix, the host automation guards and the WARP assignment path.
Every modulated control shows what its modulation is doing right now: the dry value stays put and a run in each source's colour follows the value the audio thread is using, with a combined marker where several sources meet.
Modulated filters, distortion, WARPs and oscillator previews draw a ghost of themselves at the modulated values, in the modulator's colour, on top of the dry shape. Nothing is drawn and no extra work happens when nothing is modulated.
Added analytic VA low-pass, high-pass, comb, and smoothing WARPs with source antialiasing. Added hard Sync beside Bend and PWM, including reset correction and neutral-amount continuity.
KURV 0.20 starts a new state and preset format; older project and preset formats are rejected. Preset conversion is deferred to an external migrator.
Better
WARPs use title-bar mode selection, a filled waveform on the left, and readable inside-value knobs on an LFO-style control rail. Brickwall belongs to Spectral; phase warps have their own category.
Distortion adds Amount drive alongside Mult and Mix, with horizontal and vertical skew sliders around its drawable transfer curve.
Distortion is a separate generator-stack module with a drawable transfer curve, a compact shape selector, Multiply repeats, horizontal/vertical skew, and Mix. Removed Bias and Stereo controls.
Waveshaper curve edits persist with the module and publish prepared data to audio processors only when changed; modulation uses the same transfer function shown by the editor.
Moved the envelope inside the purple group identity header. Envelope lines provide direct stage and curve editing without visible point handles.
Warp assignments use compact connection pies on the oscillator, with source-hover feedback and a centered assignment gesture on its waveform panel.
Separated general phase warps from VA harmonic low-pass/high-pass filters in the insertion menu.
Restored customized group headers and large pane labels. Group envelopes sit above the generators, with direct stage and curve editing, pointer values, and gradient fill.
WARPs use compact boxed cards with small previews, segmented controls, and a centered add button.
Compatible oscillators reveal an assignment target beside their content. Per-connection amounts stay in a compact side strip and follow host automation.
Independent WARPs rack with categorized insertion, compatible-source assignment, and resizable left and right panes. Shared warp amount multiplies each connection's own automatable/modulatable depth.
Modulation gestures now start actual drags and cancel consistently on Escape, outside clicks, and focus loss.
Added prepared Disperse/Tilt/Random and seven configurable spectral transform families, with exact zero-depth neutrality and a bounded 255-harmonic correction.
Generator-stack distortion provides six transfer types with Drive, Bias, Spread, Mix, and antiderivative antialiasing.
Group output controls follow the module chain, with a playback-derived envelope overview and responsive aligned rows. Removed the redundant group-header waveform preview.
Reduced repeated filter-coefficient bank copies; controlled dense-patch measurements preserve exact audio statistics. This stress workload still exceeds its callback deadline.
Groups own their per-note amplitude envelopes. One-group and multi-group patches use the same render path, and reordering preserves envelope and bus histories by group identity.
Source-native processors belong to their source and are edited in the left pane. Audio effects remain in the group signal path.
Admit supported native chains explicitly: one built-in VA Brickwall followed by an optional Phase Warp. Unsupported orders, source representations, spectral Warp, Tilt, Random, and Sync remain unavailable pending DSP qualification.
Faster
Recompiling a prepared WARP skips its band pass entirely when the band spans the whole spectrum, and derives the Tilt pivot once instead of once per bin. Bit-identical audio; eight voices of sixty-four unison with a 1 kHz LFO on the amount: Tilt 0.72 ms to 0.63 ms, Disperse 0.71 ms to 0.61 ms, Time Skew 0.75 ms to 0.65 ms per callback.
Per-lane harmonic limits stay in integer vector lanes instead of being cast to `usize` one lane at a time, which compiled to eight saturating scalar conversions per sample. Bit-identical audio; eight voices of sixty-four unison with a 1 kHz LFO on the amount: Brickwall high-pass 0.98 ms to 0.80 ms, Brickwall low-pass 0.76 ms to 0.70 ms, Disperse 0.74 ms to 0.71 ms per callback.
Each Brickwall prefix table row is padded so its cubic neighbourhood is four contiguous samples and its stride is no longer a power of two, so a table read is one sixteen-byte load per lane instead of four gathers into conflicting cache sets. Bit-identical audio; eight voices of sixty-four unison with a 1 kHz LFO on the amount: Brickwall high-pass 1.18 ms to 0.98 ms per callback.
A prepared-spectral table read is one sixteen-byte load per lane instead of four scalar loads, because every correction mip now repeats its first entry in a wrap pad and lane indices are derived in vector registers. Bit-identical audio; eight voices of sixty-four unison with a 1 kHz LFO on the amount: Tilt 0.96 ms to 0.76 ms, Disperse 0.94 ms to 0.74 ms, Time Skew 0.96 ms to 0.78 ms, Vocode 0.52 ms to 0.41 ms, Formant 0.53 ms to 0.41 ms, Random 0.52 ms to 0.42 ms per callback.
Brickwall derives its per-lane harmonic limits, band edge weights and prefix table positions eight lanes wide instead of one lane at a time. Bit-identical audio; eight voices of sixty-four unison with a 1 kHz LFO on the amount: Brickwall high-pass 1.48 ms to 1.18 ms, Brickwall low-pass 0.89 ms to 0.75 ms per callback.
The analytic WARP low pass skips its boundary exponential on the samples where every unison lane sits far enough into the cycle for that exponential to be under 1e-7, which is past the f32 resolution of the basis it corrects. About 2% of CURVE LP at eight voices of sixty-four unison; the output moves by about 2e-11 relative.
The phase WARPs (BEND, HARM, PWM) advance their warp harmonics by a fixed rotation seeded once per block instead of evaluating a sine polynomial per sample, and fold the displacement scale into the depth ceiling and the per-frame amounts so the block loop is a broadcast, a clamp and a fused add. Eight voices of sixty-four unison with a 1 kHz LFO on the amount: BEND 0.19 ms to 0.14 ms, HARM 0.18 ms to 0.13 ms, PWM 0.22 ms to 0.14 ms per callback. A new test bounds the rotation against the portable kernel at 2e-6.
CURVE SMOOTH answers a whole cycle from its closed form, waveform events included, whenever the smoothing width spans enough samples for the antialiasing kernel's second-moment term to be accurate there; only a cutoff above 3.6% of the sample rate still pays the per-lane quadrature. Eight voices of sixty-four unison with a 1 kHz LFO on the amount: 0.88 ms to 0.55 ms per callback. A new test bounds the substitution against a direct convolution of the ideal basis at 1e-3.
The analytic WARP filters hand the dry unison block straight to the left mix instead of writing a discarded second channel, and only add the wet difference on the frames that have one. CURVE LP 0.72 ms to 0.70 ms, CURVE SMOOTH 0.91 ms to 0.88 ms, CURVE COMB 0.57 ms to 0.56 ms per callback; the reassociated mix moves the output by about 1e-7 relative.
The Sync WARP keeps its lane batch in registers on the local-edge path instead of spilling it to arrays first. Bit-identical audio; 0.72 ms to 0.71 ms per callback.
The Sync WARP's local edge kernel takes one reciprocal per sample instead of a divide per edge family per master cycle, and each slave edge evaluates only the residual its family can carry. 0.75 ms to 0.72 ms per callback at eight voices of sixty-four unison; the output moves by about 1e-9 relative from the reassociated divides.
The Sync WARP skips its antialiasing-mode correction on the samples where no unison lane sits inside a spline kernel's edge support and the two evaluations cancel exactly. Eight voices of sixty-four unison with a 1 kHz LFO on the amount: 0.94 ms to 0.75 ms per callback. The remaining samples are unchanged; the skipped ones drop a cancelling add/subtract pair, moving the output by about 7e-9 relative.
The analytic WARP filters render their dry source with the same whole-block unison kernel the unwarped voice uses, and their eight-lane kernel resolves its filter constants once per block instead of once per sample. Eight voices of sixty-four unison with a 1 kHz LFO on the amount: CURVE LP 0.81 ms to 0.72 ms, CURVE SMOOTH 1.01 ms to 0.91 ms per callback. CURVE LP and CURVE SMOOTH are bit-identical; CURVE COMB's two shifted taps now advance as their own blocks, which moves its output by about 4e-8 relative.
the analytic WARP low pass derives its far boundary exponential from the near one instead of evaluating a second vector exponential per sample. CURVE LP on a 64-voice unison stack drops from 0.89 ms to 0.79 ms per callback, with the same audio.
analytic WARP filters fall back to the scalar antialiasing quadrature per lane instead of per eight-lane batch, so one lane whose kernel straddles a waveform event no longer costs the whole batch. CURVE SMOOTH on a 64-voice unison stack drops from 1.26 ms to 0.99 ms per callback.
Sync renders its bandlimited reset edges eight lanes wide below the primitive cancellation threshold, and skips the edge families that are out of kernel range. Eight voices of sixty-four unison with a 1 kHz LFO on the Sync amount: 2.43 ms to 0.92 ms median per callback, a 2.6x saving, with no measurable change in alias energy.
Recompiling a modulated prepared WARP only builds and stores the harmonics a correction mip can read, instead of the whole 1024-bin spectrum. Bit-identical audio; Tilt 1.00 ms to 0.94 ms per callback.
The prepared-spectral correction lookup derives its table indices eight lanes wide, from mip geometry resolved once per block, instead of one lane at a time. Bit-identical audio; Vocode 0.75 ms to 0.53 ms, Tilt 1.23 ms to 1.00 ms per callback at eight voices of sixty-four unison.
A modulated prepared WARP rebuilds its spectrum only on the samples that can compile a new endpoint, instead of once per sample and discarding sixty-three of every sixty-four. Bit-identical audio; Vocode 0.91 ms to 0.75 ms, Tilt 1.32 ms to 1.23 ms per callback at eight voices of sixty-four unison.
Prepared-spectral WARPs (Spectral, Tilt, Disperse, Time Skew, Random) share one table index across the crossfade endpoints and choose their correction mips once per block instead of once per sample. Bit-identical audio; eight voices of sixty-four unison with a 1 kHz LFO on the amount: Tilt 2.13 ms to 1.32 ms, Disperse 2.10 ms to 1.33 ms, Time Skew 2.10 ms to 1.32 ms, Vocode 1.11 ms to 0.91 ms per callback.
The block-major WARP amount path is no longer limited to single-group patches, so cross-group audio-rate modulation keeps eight-lane block rendering. Eight voices of sixty-four unison with a 64-lane oscillator on a WARP amount in another group: 2.18 ms to 0.24 ms median per callback, no missed deadlines.
Every WARP renders block-major and eight lanes wide, including Brickwall, Tilt, Disperse and Time Skew, which previously dropped the whole voice back to sample-by-sample rendering. Modulated WARPs on sixty-four unison voices cost between two and five times less CPU. x86 builds now require SSE4.1.
Fixed
Audio-rate sources now cross group boundaries in both directions. An oscillator in one group can drive a WARP amount, a phase, or a pitch in any other group, in either direction and in mutual pairs; the render order is a topological sort of the whole oscillator graph, and a cycle reads the previous sample. Previously the audio source had to precede its targets inside the same group.
Oscillator previews include assigned warps and independent connection amounts. Connection markers expand with the oscillator rail, show names only while dragging, and double-click removes the assignment.
Collapsed envelope stages remain draggable to increase their duration. Strengthened the envelope fill and removed its bottom baseline.
Apply source gains once in block rendering and retain per-voice Group Pitch modulation. Save and restore allocated host values through the same configuration boundary.
Keep plugin creation on a bounded host stack by constructing the boxed DSP state on a dedicated initialization thread.
Rebuilt Settings around Input, Capture, Export, Appearance, and Performance, with License, Updates, Diagnostics, and Report inside one window. Stable accessible names for icon-only controls, consistent spacing, stronger headings, aligned icons, solid accents, readable slider values and FFT menus that close after selection, and compact native dialogs replace nested panels and uneven padding. Theme and spectral colors edit directly, save automatically, and reset to their original presets. Settings preserve valid saved choices across upgrades and merge changes between instances; new configurations default to icon-only toolbars, solid waveform fill, and two hours of history.
Added faster GPU spectral analysis and wavelet synchrosqueezing for clearer separation of nearby notes, plus GPU spectral selection where supported and CPU fallback. Faster parallel loading, reusable zoom detail, adaptive memory limits, Overview and Full detail modes, corrected frequency scaling and tile edges, stable selections, stronger narrow ridges, calibrated levels, and a wider Floor range improve analysis and display. Long recordings, compressed imports, processed previews, playback, exports, and undo use bounded storage; recovery survives plugin reloads, restores faster, and preserves unreadable data safely.
Improved recording and playback reliability: System Audio continues when the host stops processing, macOS capture handles formats and timing more reliably, and source changes and MIDI preview avoid interrupting other channels. Corrected quiet multichannel playback, true-peak normalization, streamed processing continuity, silent-range exports, Windows drag-and-drop, oversized host blocks, and shutdown behavior. Graphics setup, lost devices, Spectral reopening, and auxiliary-window commands recover more reliably, while background work avoids blocking recording.
Setup starts with activation, adds a final Make it yours step for theme, waveform, and spectrogram choices, and finishes with a three-second BUFFR-colored inward animation, animated logo, and reduced-motion support. Activate by key, start a trial, or use MuseHub in supported editions directly inside Setup or Settings. The License page shows a masked key, current device and active devices, with refresh and confirmed deactivation. Existing activations are preserved.
Preview is consistently named in updates and downloads, with installed and available release notes and scrollable history. Diagnostics show the complete available session log. Reports always include viewable diagnostics; the displayed snapshot is the one sent. Automatic crash reporting retains complete available incident evidence until delivery is confirmed, and more accurate graphics, memory, storage, and shutdown diagnostics improve troubleshooting.
Universal macOS installer ZIP (.pkg inside) · Windows x64 (.zip) · Linux x64 (.zip)
Simplified editor and aligned responsive layout. Fixed CLAP state refresh, VST3 host interface discovery, and clipped rendering on resize. Windows and Linux CLAP/VST3 builds; Windows requires Microsoft Visual C++ 2015–2022 x64 runtime. Run the included Windows installer as administrator and restart Windows before scanning plugins.
Windows x64 (unsigned) · Linux x86_64 (Ubuntu 24.04+)
Interactive F1 help. Hover any component for an outline, press F1 or click HELP to lift it out of the workspace, enlarged over the dimmed background and still fully live; callouts name every control with its range and meaning.
A full manual for every module. The Index at the top right opens pages for the VA oscillator, wave drawing and VA tables, waveform expressions, Noise, Grain, Resynth, unison, pan, filters, envelopes, LFOs, macros, modulation routing, performance and MPE, groups and audio routing, presets and settings, each with control-by-control tours, gestures, limits and recipes.
Help navigation: ↑ and ↓ walk the controls of the staged component, ← and → step to the neighbouring component, Enter drills into a region, Backspace climbs back out, Escape closes.
MACROS page and a MacroPack section for macro knobs, ON/OFF buttons and the XY pad, with F1 on those cells opening the right page.
Oscillator LEVEL now reaches 200% instead of 100%, on the VA readouts, the Noise card and audio-rate AM routes.
Armed modulation mode keeps every control editable: click a source port to arm it, press the + on each destination to route and set its depth, and keep adjusting values in between.
Group PITCH transposes every pitched oscillator in a group by up to ±48 semitones, and each modulator has a bypass that is saved with the patch.
A rebuilt card shell shared by every rack card (oscillators, filters, AUX, sources and group headers), with one readout vocabulary, a six-dot reorder grip on every card and the same fine, snap and reset gestures everywhere.
Better
The help backdrop is a plain dim veil instead of a blur, so the workspace behind the stage stays intact.
F1 over any part of a card (its header, a graph, a cell) lifts the whole card rather than a single region.
Every host parameter, the settings panel, the preset deck, the wave editor pencil, function toggle and table actions, and the quality selectors carry help text.
Filter cutoff, resonance, slope, morph and shape modulation now cover their real ranges, and filter mode changes stay continuous instead of clicking.
The OBJECT filter's modes are peak-normalised, so DECAY no longer doubles as a volume control and nothing comes out louder than the input.
The group envelope callouts read their true 0 ms to 20 s range and the pitch wheel range reads 1 to 96 semitones.
Waveform expression edits are bounded and can no longer overflow the compiled buffer; invalid edits keep the last valid wave playing.
The stereo pair of a diffused Rich resynthesis is now decorrelated instead of duplicated.
Faster
Editor parameter block shrinks from about 500 KiB to 76 KiB and the host automation pool from 2048 to 512 slots, with deleted-module slots kept reserved until the pool is full.
DSP state at load drops from 125 MB to 41 MB by allocating Resynth voice state only for slots that hold a sample.
Prepared SVF, Scream, Phaser, Object, oscillator and structural render state is reused on hot paths, with 2 to 15% lower median callback time on dense, feedback and modulation-heavy workloads.
Zero-attack envelopes are declicked with a band-limited step and passband equalisation keeps 1x, 2x, 3x and 4x oversampling tonally matched.
Worker calibration waits are bounded so an audio callback can never be extended by a background job.
Fixed
Embedded Resynth samples no longer go silent after a Bitwig deactivate and reactivate cycle or a project restore; Grain, Rich and sample slots are republished on every artifact publication.
The LFO graph keeps responding while its card is lifted onto the help stage.
No remaining panics on the audio thread; the whole crate now passes clippy with a baseline of a handful of warnings, down from over six thousand.
The test suite is green (418 tests) after fixing block-renderer divergence during gain declick windows and two timing-dependent pool tests.
Audio and MIDI imports append cleanly with mixed sample-rate conversion, complete undo state, bounded multichannel disk use, visible progress, correct keyboard focus, and immediate waveforms; repeated exports reuse prepared files and remain reliable through Windows Explorer, asynchronous hosts, Wayland, XWayland, undo, cancellation, and editor closure.
Settings, Activity Log, Diagnostics, and Report panels are movable, resizable, scrollable, selectable, copyable, and accessible; reports retain bounded capture, renderer, Spectral, memory, import, and crash evidence, build IDs identify exact installs, damaged trials recover, and activation, support, and update downloads reject malformed responses.
Faster
Long recordings keep sealed audio on durable storage, release obsolete audio and Offline FX previews, bound waveform, Spectral, and MIDI memory plus multichannel undo disk use, and avoid repeated scans, scene copies, gradient work, idle redraws, and background-worker polling; Auto, Full, and Low Power modes adapt presentation cost to the machine.
Fixed
Native drag, resize, input, focus, modal, teardown, and late host callbacks no longer freeze, crash, abort, or retain destroyed editors; Baseview 0.3.2 improves CLAP, VST3, standalone, multi-instance macOS Metal, software-rendered Windows, embedded Linux, X11 parent detection, reopening, resizing, and display scaling.
System Audio follows device and host changes, survives uneven processing and recoverable interruptions, reports startup failures and underruns accurately, and guides macOS users to Screen Recording settings; standalone shutdown, shared MIDI routing, controller reconnection, per-note expression, and accessible MIDI activity are more reliable.
Universal macOS installer ZIP (.pkg inside) · Windows x64 (.zip) · Linux x64 (.zip)
BUFFR-V now ships as a separate experimental CLAP, VST3, and standalone download for Linux, Windows, and universal macOS, with its own plug-in identity and notarized macOS installer.
The experimental Vizia spectrogram now retains source tiles while the timeline moves and colors them on the GPU, removing the viewport-wide CPU recoloring that stalled playback and resizing.
Completed spectral analysis now redraws the Audio canvas immediately, and playback schedules spectrogram work ahead of the moving playhead instead of waiting for unrelated UI activity.
Spectrogram redraws now target the correct editor and stop cleanly when it closes, preventing stalled multi-instance views and redundant frame requests.
Partial spectral tiles no longer sample an empty neighboring column while recording or panning between analysis columns.
Universal macOS installer ZIP (.pkg inside) · Windows x64 (.zip) · Linux x64 (.zip)
RATIO spectral filter mode masks an oscillator's partials at the source instead of the mixed bus: CUTOFF is a harmonic ratio from 0x to 1024x, PASS flips between harmonic low-pass and high-pass, and the filter panel previews the masked wave.
Settings is split into INTERFACE, WORKSPACE and the remaining pages with a category selector instead of one long scroll.
UI SCALE setting with 100%, 125% and 150%, saved with the plugin state and composed on top of the host DPI.
MACROPACK source card in the LFO rack with up to 12 macro cells, each assignable to several parameters, with BUTTON toggle cells and unipolar or bipolar switching.
Per-parameter undo and redo: hover a control and press Alt+Z or Shift+Ctrl+Z to step only that parameter's history, next to the whole-plugin undo.
Performance panel gains separate PB RANGE down and up values (1 to 96 semitones), an MPE RANGE control and a typed TUNING entry in Hz.
Group cards can be renamed and recoloured from their identity chip, dragged to move, Ctrl-dragged to duplicate, and paint a live cycle preview of the group output including its modulation routes.
AUX slots get a compact card with a drop zone for audio-rate oscillator sources, LEVEL, INVERT, and reset and remove actions.
Better
The route inspector shows destination and depth inline, and a source can be flipped between unipolar and bipolar from the inspector.
Dragging a source onto a target explains itself: vertical drag sets the depth, and invalid targets are named as such instead of silently refusing.
Ctrl snaps parameter drags to musically meaningful steps (semitones, cents, percent) on top of Shift for fine control and double-click to reset.
Oscillator cards show the tuning mode (SEMI, RATIO, EXP FM) as a clickable switch with matching cents, Hz and ratio readouts.
Pan-shape and spline editors: double-click a point to flatten its segment, double-click empty space to add a point, Shift for fine moves, remove and reset in the right-click menu.
New settings for modulation display and CPU: SHOW MODULATION CABLES keeps cables visible for every visible route, and AUDIO-RATE FAST or EXACT trades CPU against modulation quality.
Faster
Modulation DSP and generator routing reworked and measured against realtime budgets across multi-group, AUX, audio-rate FM/PM/AM/ring and resynth workloads.
macOS Universal (.pkg) · Windows x64 (.zip) · Linux x86-64 (.tar.gz)
Auto, Full, and Low Power display modes now let BUFFR adapt to constrained machines while preserving the user's full-quality settings.
Better
Completed audio imports reveal the full appended waveform immediately, while the animated aurora keeps the lane visibly busy until the import finishes.
Audio imports now keep the waveform visible under an animated aurora until ready, block conflicting gestures during the handoff, and report the exact reason when a drag export cannot be prepared.
Timeline pan and vertical gain, frequency, and pitch controls now expose distinct values and actions to assistive technology.
Long recordings keep using conservative waveform summaries at zoom boundaries instead of redundantly rereading raw audio.
The editor stays idle when nothing changes, coalesces redraws during active work, and preserves native drag, accessibility, and graphics-recovery behavior.
Panel close buttons now expose their purpose to screen readers and desktop automation.
File-drop diagnostics now record background import progress so stalled imports can be identified without guessing whether the host accepted the drop.
Support reports include bounded current activity, correlate delayed crash artifacts more reliably, and finish background submissions safely before the host unloads BUFFR.
Auto, Full, and Low Power display modes now adapt to constrained, virtual, and software-rendered machines while preserving Full settings; Low Power uses a one-second passive refresh, a simple waveform, and disables Spectrogram and background Wet work, with performance controls available from FX.
Standalone closes without waiting for another audio callback, including before input starts and when JACK stops; MIDI activity is announced accessibly, linked inputs reconnect more reliably, and per-note expression stays attached to the correct repeated note; macOS crash reports retain memory-allocation evidence and support reports can be correlated anonymously.
Standalone closes without waiting for another audio callback; MIDI activity is announced accessibly, linked inputs reconnect more reliably, and per-note expression stays attached to the correct repeated note; macOS crash reports retain memory-allocation evidence and support reports can be correlated anonymously.
MIDI activity is announced accessibly, linked inputs reconnect more reliably, per-note expression stays attached to the correct repeated note, and Low Power bounds dense note and marker drawing; macOS crash reports retain memory-allocation evidence and support reports can be correlated anonymously.
macOS crash reports retain virtual-memory allocation and memory-summary evidence for diagnosis.
Crash reports can now be correlated with anonymous installation and system statistics without exposing raw hardware identifiers.
Right-clicking FX now opens performance, refresh-rate, Spectrogram resource, and background Wet controls.
Faster
Import undo history stores only channels containing retained audio instead of every negotiated host channel, greatly reducing temporary disk use in multichannel hosts.
Long recordings now keep sealed audio on durable storage instead of retaining the complete capture in memory; clearing, replacing, or changing Offline FX releases obsolete resident audio and waveform summaries.
Zoomed-out waveforms avoid repeated recording scans, scene copies, gradient work, and continuous transient redraws; Low Power bounds dense MIDI note, marker, and expression memory while long-session paging skips empty slots.
Zoomed-out waveforms avoid repeated recording scans, scene copies, gradient work, and continuous transient redraws; long-session paging skips empty slots, and background workers stop without idle polling delays.
Software-rendered editors now select Low Power automatically and limit passive live animations and waveform refresh without slowing direct interaction.
Zoomed-out waveforms avoid repeated scans, scene copies, and continuous onset redraws; long-session paging skips empty slots; background workers stop without idle polling delays.
Zoomed-out waveforms no longer rescan the visible recording for every drawn bucket.
Low Power uses a 30 FPS simple waveform, hides transient and fill work, disables background Wet preparation, and keeps Spectral processing inactive.
Waveform redraws now reuse display calculations, skip unused gradient work, and enforce the selected presentation rate for background-triggered frames.
Spectral processing now starts only when Spectral view is used, and switching back to Waveform releases its cached analysis.
Long recordings now keep sealed audio on durable storage instead of retaining the complete capture in memory, while preserving playback, waveform, effects, and export access.
Repeated Offline FX preview changes now release obsolete waveform summaries instead of accumulating display memory.
Fixed
Audio files with different sample rates are converted to the active timeline rate while importing, so repeated mixed-rate drops append, display, play, and export as one continuous buffer.
Timeline keyboard shortcuts keep working after clicking the audio or MIDI lane while text fields still retain keyboard input.
Export drags on Windows now keep their native thumbnail after leaving BUFFR so it follows the cursor through Explorer and compatible hosts.
File drops from native Wayland applications reach BUFFR again when it is embedded in XWayland DAW windows, with an incoming-drop indicator before release.
Export drags work after importing audio whose sample rate differs from the DAW sample rate.
Embedded Linux editors no longer crash when a host window rejects a focus request.
Rebuilt the editor on Baseview 0.3.2, kept Settings and Diagnostics inside the main window, and hardened resize and input handling against reported editor crashes.
Editor opening, closing, reopening, resizing, and UI scaling now follow the host window contract consistently across CLAP and VST3.
Late host callbacks after an editor closes are ignored safely instead of retaining or re-entering a destroyed editor.
Windows standalone now uses the same native graphics fallback as hosted formats instead of failing on systems without working OpenGL.
Activation, support, and update downloads now reject oversized or malformed responses cleanly.
Windows waveform selections no longer collapse when the window backend repeats a drag-start event.
Dragging an exported selection into Windows Explorer no longer freezes BUFFR during the native file handoff.
Linux X11 resize handling now distinguishes actual DAW parents from window-manager decoration frames, addressing the observed standalone runaway resize and memory growth while preserving embedded resizing.
Dragged audio, MIDI, and combined files remain available for DAWs that finish importing asynchronously after a cancelled, rejected, or failed native handoff; repeated audio drags keep reusing the prepared file.
Diagnostic storage failures and unavailable snapshots are now reported explicitly instead of appearing successful or empty.
Repeated drags reuse prepared files, successful or ambiguous handoffs survive undo and editor closure, and rejected or failed drags release temporary files; native drag, resize, and modal callbacks no longer re-enter rendering, and Windows System Audio follows output changes and closes cleanly during host teardown.
Windows System Audio now closes cleanly when a host validates or unloads BUFFR while continuing to follow default-output changes.
System Audio status now refreshes immediately after startup or shutdown, and recoverable capture discontinuities no longer restart capture.
MIDI input controls now announce live activity to screen readers.
Repeated audio and spectral drags now reuse one prepared file, and files already handed to a host survive cancellation, undo, and closing the editor.
Nested native window callbacks no longer freeze or crash BUFFR during file dragging, resizing, or modal window interactions.
Plug-in teardown no longer aborts when the final background worker finishes during shutdown.
Parked timeline zoom and file dragging now respond immediately instead of waiting for background repaint limits.
Windows plug-in and standalone interfaces no longer freeze or crash when native input arrives during a redraw.
Clearing or replacing a recording now releases its resident audio memory without discarding retained audio.
Universal macOS installer ZIP (.pkg inside) · Windows x64 (.zip) · Linux x64 (.zip)
First-run Capture setup now lets users choose whether offline processing is cached in the background, with clear memory guidance on hover.
With background Wet preparation disabled, selecting audio starts one exact preview with visible progress; enabling it also prepares reusable full-buffer previews.
Updating on Windows now requests permission, waits for the host to release BUFFR, replaces the standard CLAP and VST3 installs, and places the standalone app on the Desktop.
Updating on macOS now opens the signed, notarized BUFFR installer in Apple Installer.
Linux updates place the standalone app on the Desktop and request administrator permission only when an existing system-wide plug-in copy also needs updating.
First-run setup now uses two focused Capture and MIDI steps, with format help on hover and Bit depth beside History.
Global MIDI source and multi-controller linking now share one setup page with a clearer alternating device list.
Finishing setup now blooms into a brief BUFFR-colored Ready moment before revealing the recorder.
Dropped audio and MIDI now append at the end of their lanes, and one undo restores the complete pre-drop audio and MIDI state, including combined drops.
Audio drops target the selected channel, while All preserves BUFFR source-channel routing when available; imported MIDI keeps its embedded MIDI channels.
The drop target now highlights the affected lane and audio destination with a foreground dashed aurora overlay.
Linux installation guidance now lists the runtime libraries and graphics support required by the release package.
Platform guidance now states the Windows MIDI Services, Linux runtime-library, and macOS deployment baselines required for a reliable BUFFR install.
Global MIDI now relies on native sharing on current Windows 11 systems. Exclusive legacy Windows inputs should use Host MIDI from the DAW.
Extended plug-in validation now allows enough time for large multi-rate Windows validation runs to complete.
Settings now exposes locally published experimental releases as the Unstable update channel.
The current Nice Plug and Baseview editor lifecycle improves native window ownership, display scaling, and editor open/close behavior across CLAP, VST3, and standalone builds.
BUFFR now uses the current upstream egui rendering and input backend while preserving accessibility, native drag-and-drop, frame scheduling, and graphics recovery behavior.
Keyboard input uses the current cross-platform event model while preserving BUFFR's native window, display-scaling, and drag behavior.
Offline effect-analysis caches now rebuild safely from retained recordings after the experimental dependency update.
Experimental builds refresh the audio, graphics, resampling, analysis, and updater foundations for current platform compatibility.
Faster
Background Wet analysis now has a fixed memory budget, while disabling it releases the cache and avoids duplicate selection export rendering.
Silent rolling capture now leaves the interface truly idle instead of continuously redrawing an empty timeline.
Unchanged interface frames now reuse the last presented image instead of rebuilding and presenting the complete window.
Spectral analysis now stops at the newest safe FFT window instead of repeatedly scheduling impossible future work.
Software-rendered editors limit live redraws, and background effect analysis no longer busy-spins while recording.
Audio processing now flushes denormal values to prevent avoidable CPU spikes on x86 systems.
Spectrogram rendering reuses one GPU-coverage snapshot per frame and avoids rescanning unchanged tile memory while idle.
Fixed
Wet waveform previews now require exact continuous cache coverage, and unavailable audio can no longer appear as a valid silent Wet signal.
Successful drags from an XWayland BUFFR editor into Bitwig now release the source gesture cleanly, so later drags work without reopening the interface.
New audio activity wakes the waveform immediately, while reopening or refocusing the editor always restores a complete frame.
Back always returns to Capture, and completed steps can be reopened from the setup progress bar.
Files dragged from another BUFFR window or a Wayland app now reach embedded BUFFR editors on Hyprland.
A failed Hyprland drop no longer blocks later drags until the editor is reopened.
Windows software-rendered editors now stay responsive during timeline selection, settings changes, and live capture without returning to excessive redraw load.
Spectral view now fills the complete visible recording during live capture instead of leaving thin analyzed chunks separated by gaps.
Rapid Windows editor automation and resize callbacks no longer crash the host while a frame is rendering.
Standalone audio and MIDI now work with interfaces that require native input or output buffer sizes, including MOTU M2 capture and Windows WASAPI devices.
Windows hosts can scan, instantiate, unload, and reopen saved BUFFR projects reliably while Global MIDI detects device changes in the background.
Rapid Windows editor automation and resizing no longer crashes the host or overwrites the user's interface preferences.
The standalone window now displays the BUFFR title instead of a framework placeholder.
Standalone MIDI input now processes complete controller bursts per audio block, preventing chords and dense control changes from being delayed across later buffers.
The macOS AU now targets macOS 13 instead of inheriting macOS 15 from the release builder.
Linux Global MIDI now verifies the exact controller endpoint and live owner before offering one-time RawMIDI compatibility setup, avoids incorrect setup prompts, and reports the saved supervised controller in route status.
Updating BUFFR now restarts an active Linux MIDI route service so it immediately runs the newly installed helper.
Rapid editor close and reopen cycles no longer abort Linux hosts while the graphics surface is being recreated.
VST3 parameter automation no longer wakes an operating-system queue from the real-time audio thread, reducing intermittent dropout and stability risk.
Linux only offers its one-time shared MIDI setup when another application actually holds the selected controller.
Audio and MIDI activity now refreshes the editor promptly without waking background services from the real-time audio thread.
VST3 host bypass controls now stay synchronized while audio processing is active and after state restoration.
Rapid host automation of UI scale no longer compounds window dimensions or stalls the editor on Linux.
Repeated Windows plug-in state save and restore no longer crashes after Global MIDI device monitoring starts.
Windows builds work with the upgraded accessibility and native file-drag backends.
Audio and MIDI activity now wakes the editor immediately, while text, icons, and controls stay visible and correctly scaled across displays and zoom levels.
Corrupt plug-in state is rejected without crashing; restored values, saved defaults, and audio-buffer state remain consistent during host state and lifecycle handoffs.
Host deactivation now cancels offline effect analysis promptly instead of waiting for obsolete catch-up work.
Native Wayland editors can be reopened reliably, shut down cleanly when the plug-in unloads, and no longer report a file drag as complete before the destination finishes reading it.
The BUFFR logo renders correctly again throughout the experimental graphics build.
Spectrogram analysis now reaches the GPU continuously instead of appearing as thin chunks that slowly fill while zooming.
The standalone app now keeps running through recoverable audio-device changes, stream dropouts, and denied real-time priority instead of treating them as fatal errors.
Universal macOS installer ZIP (.pkg inside) · Windows x64 (.zip) · Linux x64 (.zip)
Rebuilds KORREKT as a minimal Center, Punch, and Spectral instrument, with the live oscilloscope inside the giant draggable Center control and real values inside both smaller controls.
Auto compares identity, a conservative bank, and the full spectral search, then replays candidates through the shipped lattice filter before publishing the safest winner.
Adds four real analysis lookbacks with bounded search budgets, adaptive click-free pole transitions, a measured spectral rainbow, and visible zero host latency.
Punch now uses Hann-windowed harmonic detection and must beat a focus-free Auto finalist under exact continuous lattice replay before it can replace the safe centered bank.
Faster
Pole changes now finish through finite staged all-pass paths, coalesce rapid analysis updates to the newest target, and measure their adaptive rate against retained programme audio.
Programme-level fallback refinement searches nearby frequency and radius basins, while deep Punch candidates must sharpen four unseen analysis windows without sacrificing centering or true peak.
Gives the editor a sport-orange visual system with neutral charcoal surfaces, athletic Barlow Semi Condensed typography, sharper geometry, lime telemetry, and a vibrant multicolor spectral palette.
Restores the rotary Amount, Depth, and Focus controls with refined instrument styling, keeps precision sliders for time-based controls, and replaces tinted surfaces with a neutral charcoal palette.
The editor is rebuilt as a focused phase-analysis instrument with custom precision rails, flat mode matrices, clearer telemetry, Archivo typography, and a restrained graphite-and-ice visual system.