A native, open-source DSP rack and per-app router for macOS system audio. No virtual audio driver for output; an optional one for the microphone.
- Modular effects chain on all system audio, reorderable by drag, up to 16 modules:
- Tone: Gain, Parametric EQ (up to 32 bands: bell, shelves, pass, notch), 16-band Graphic EQ, Filter (12–48 dB/oct), Loudness
- Character: Bass enhancer, Exciter, Crystalizer, Crusher
- Dynamics: Compressor, Expander, Gate, De-esser, Limiter, Maximizer, Autogain
- Space: Stereo tools, Crossfeed, Delay, Reverb (Freeverb)
- Contextual chain placement: a new module lands where it belongs in the signal path, after the last module of an earlier or equal stage. The order is gain staging → gate/expander → de-esser → filter/EQ → compressor/autogain → saturation → stereo/time effects → loudness → maximizer → limiter. Existing modules are never moved; drag if you want something else. A new EQ filter is inserted at the top of the list.
- Headphone correction from two sources, applied as a parametric module in one click: 8,800+ profiles from jaakkopasanen/AutoEq (oratory1990, crinacle, Rtings, …), or any reviewer database on squig.link — pick the database and target curve, and the correction is computed from the measurement on the spot. Hangout.Audio (crinacle's own site) opted out of third-party access and is not offered; his measurements reach the app through AutoEq.
- Any filter count: AutoEq publishes ten filters per profile; ask for 5 to 32 and patchbay fits its own from the full-resolution correction (on the HD 650 the 10-filter fit lands within 0.25 dB RMS of the target, better than the published ten; 16 filters halve that).
- Equalizer APO import/export:
ParametricEQ.txtin,ParametricEQ.txtout - Per-device chains: every output device remembers its own rack
- Presets: save the chain under a name and recall it on any device, route or the microphone. A preset saved from an output is remembered for it, so the same headphones on a fresh profile (or a new pair with the same name) start from that chain automatically.
- Bypass for instant A/B, input/output metering, device sample rate picker in the rack footer
- Output and input device switching, hardware volume, mic gain and hardware mic mute
- eqMac recovery tools (fix stuck audio, restart, reset Core Audio) on the Fix page
- App routing: send one app to a different output device, with its own
chain. Everything not routed keeps using the system chain. Helper processes
(browser renderers, Electron utilities) follow their app; command-line
players (
afplay,mpv,ffplay) are routed by executable name.
One menu bar popover with five icon tabs: output, input, routes, rack, fix. It opens and closes without animation; in-app motion is short springs.
- Layout follows the page by default (Auto): one width, device pages compact and only as tall as their content, the rack spacious and capped at 640 pt with the module editor scrolling inside. Pages crossfade; nothing is rebuilt on a tab switch. Compact, Comfortable and Spacious pin one density everywhere.
- Settings live behind the gear in the footer: appearance (system, dark, light — applied to the popover itself), layout, Stay open (otherwise a click anywhere else closes the window), the shift step, accent colour, audio capture topology, and the virtual microphone driver.
- The chain is a vertical column beside the module editor in signal order, first stage on top. Click a row to edit it, press and drag to reorder (the row follows the pointer, the others slide out of its way), dot to bypass one module. The sidebar button in the rack footer hides the column; the editor's title then becomes the module picker, with Add inside it. The icon beside the name in the header is the chain being edited: waveform for the system chain, branch for a route, mic for the microphone.
- The preset chip under the chain header shows which preset the chain came from, with a dot once you have edited past it. Its menu applies a preset, saves the chain as a new one, updates, renames or deletes the current one, and toggles Auto-select for . With the column hidden the same items live in the module picker's Preset submenu.
- Sliders: hold shift while dragging for fine control. The drag turns relative and slows to the Shift step set in Settings (½ to 1⁄50 of pointer travel), and a bubble on the knob shows the value being dialled. Applies to every slider, including the EQ gain columns.
- Parametric EQ is a row of vertical gain faders, one per filter, low to high frequency. Tap a column to edit its type, frequency and Q in the rows below; the currently selected column is highlighted. Once the columns no longer fit the editor, the row scrolls sideways. The chips under the search field pick the source (AutoEq or a squig.link database), the filter count, and for squig.link the target curve to correct towards.
- Graph (the pulse button in the rack footer) toggles a panel showing the combined frequency response of every enabled linear module (EQs, filters, loudness, gain) as a solid curve, the selected module's own curve dashed when it differs, and a live spectrum of the processed output behind them. Dynamics, saturation and space modules have no fixed response and are not drawn. The analyser only runs while the panel is visible.
- Routes lists each app → device pair with a live status dot (waiting for
the app, processing, error). The
+menu offers every app currently connected to Core Audio. The sliders button on a row opens that route's chain in the rack; a scope chip at the top of the chain column switches between the system chain and each route. The header switch and footer status refer to the route being edited only while the rack page is in front.
patchbay uses the Core Audio process-tap API (macOS 14.2+) instead of installing a virtual audio driver:
system audio (minus routed apps) → system tap → [ chain ] → default device
app A → route tap → [ chain ] → device X
Each tap and its output device are combined into a private aggregate device.
An IO proc on that aggregate reads the tapped mix, runs the chain, and writes
the result to the hardware. A route's tap is a stereo mixdown of exactly that
app's processes; the system tap excludes them so nothing is captured twice.
When an app's process set changes, the live tap's description is updated in
place (kAudioTapPropertyDescription); the pipeline is only rebuilt if the
HAL refuses. Nothing is installed into the system.
Safety properties of this design:
- Capture is proven before muting. The tap starts unmuted; only after real samples arrive and the output layout is confirmed does the engine rebuild with a muted tap. A denied permission or unsupported device cannot silence the Mac.
- Crashes cannot strand audio. If patchbay dies, coreaudiod destroys the private tap and aggregate and the real device remains the default output.
- Realtime path is allocation-free and lock-free. The UI publishes immutable
config snapshots through an atomic pointer (
DSPConfig.c); filter memory is kept across parameter changes so slider moves never click. Output is NaN-guarded. - Routes prove capture once. A route whose app goes quiet keeps its proof, so the app's next sound is captured muted immediately instead of leaking a proof window to the default device.
Tap topology is chosen in Settings → Audio capture. Stereo mixdown (default) has Core Audio mix every process to one stereo stream in its own format, which patchbay processes and the aggregate resamples to the device rate when they differ; it works on every device. Device stream binds the tap to the output device's hardware stream, so the format matches exactly and nothing is resampled; cleaner on paper, but some devices deliver silence, so it is opt-in.
- Input effects need a virtual device. Process taps only intercept output. Putting the chain on a microphone so other apps receive it needs something that looks like an input device, which is why the microphone chain is the one opt-in that installs a driver (see Microphone below). Without it, input device selection, gain and mute still work.
- One tap processor at a time. Running patchbay alongside another
mutedWhenTapped-based app (FineTune, CoreEQ) chains two muting processors on one device; results range from double latency to silence. - Not ported from EasyEffects: convolver, multiband compressor/gate, pitch, RNNoise/DeepFilterNet noise reduction, echo cancellation, speech processor.
Requires macOS 15+ (Apple Silicon) and the Xcode Command Line Tools.
./build.sh
open patchbay.appOn first rack enable, macOS asks for System Audio Recording permission. The rack is deliberately off at launch: patchbay never seizes system audio without being asked.
The one feature that touches the system. Settings → Virtual microphone → Install
copies a 90 KB passthrough loopback driver into
/Library/Audio/Plug-Ins/HAL/patchbayMic.driver (asks for an administrator
password, restarts Core Audio for about three seconds). The driver is
BlackHole (GPLv3, Existential
Audio) built as two devices sharing one ring buffer: "patchbay Mic", visible and
input-only, which apps select; and a hidden output-only sink the engine writes
into, so the engine can never read back its own output. The source and
build script are in VirtualMic/. It has no logic of its own: whatever is
written to its output stream appears on its input stream.
real microphone → [ microphone chain ] → patchbay Mic → Zoom, Discord, OBS…
With the driver installed, the switch on the Input page runs the chain from the selected real microphone into patchbay Mic and makes patchbay Mic the default input, so apps pick it up without configuration. Turning it off hands the default input back to the real microphone. The chain is edited in the rack via the scope chip (Microphone) and is remembered per microphone.
On quit, and whenever the engine cannot run (driver missing, no microphone), patchbay hands the default input back to the real microphone, so apps are not left listening to a silent device. That covers Quit, logout and SIGTERM; after a crash the hand-back happens on the next launch. Expect roughly 10 ms of added latency. Remove in Settings deletes the driver and restarts Core Audio again.
GPLv3. AutoEq data is MIT-licensed by Jaakko Pasanen and contributors. The virtual microphone driver is BlackHole, GPLv3 © Existential Audio Inc. (VirtualMic/LICENSE).