0
Discrete voices
Eight separate circuits, never a shared DSP pool. Voices drift apart by ±0.4 cent. That drift is the chorus.
Thermal drift
Warm-up takes eleven minutes. The pitch settles, then breathes.
OSCILLA
An eight-voice analogue instrument built for rooms that remember.
No screen. No menus. Copper, current, and the two milliseconds between intention and sound.
1,248 hand-soldered parts
0.4 ms MIDI 2.0
512 patch memories
Play the machine
Scroll to begin
New · MK-VIII Rev 4Discrete VCA per voiceClass-A ladder filterMade in Leipzig240 units only
01 — Manifest
We removed the screen, the menu and the subscription. What survives is copper, current and the two milliseconds between your intention and the room's answer. Everything else was noise.
0
Independent analogue voices, each with its own filter, amp and slight, useful flaw.
0
Components soldered, matched and drift-tested by four people in one workshop.
0
Instruments this year. The number is a constraint, not a marketing decision.
02 — The instrument
Touch it. It answers.
This is the actual engine, running in your browser. Turn the knobs, play the keys with ASDF GHJKL, or run the sixteen-step sequencer. Headphones recommended — the reverb tail is long.
OSCILLA MK-VIII · ENGINE ONLINE
Standby — click anywhere to wake the audio core
Transport
ASDF… play keys · Space transport · click a voice below to re-patch
C3Two octaves · equal temperament · 441 Hz referenceC5
Ladder filter, self-oscillating above Q 17Rebuildable firmwareTen-year warranty, parts for thirtyNo telemetry, ever
03 — System
Twelve decisions
Every choice in the MK-VIII cost us something. Here is the ledger.
“It is the only machine I own that sounds better when I make a mistake.”
Ada Kestrel · Composer, Nine Winters
/02
Mod matrix
32 slots, any source to any destination, sampled at 48 kHz.
0.4 ms
MIDI 2.0 with per-note expression, plus a genuine analogue CV path in and out — because some pedals predate the protocol.
Serviceable forever
Through-hole boards, published schematics, no potting compound. Every screw is a standard size. We would rather you repair it than replace it.
/06
The room is the ninth voice
A convolution stage built from forty impulses captured in concrete stairwells, drying barns and one decommissioned reservoir in Saxony. Two seconds of tail, no plugins.
Barn · 1.9 sStairwell · 1.2 sReservoir · 4.6 sChapel · 2.8 sTunnel · 3.4 s
Zero telemetry
No account, no cloud, no firmware phone-home. The only thing this instrument transmits is sound.
04 — Signal path
From finger
01 / 05
From finger
to air
STEP 01
Contact
The key is a velocity sensor under a weighted wooden lever. 0.4 mm of travel registers between 0 and 127. There is no aftertouch because we could not make it feel honest.
STEP 02
Generation
Two saw-core oscillators and a sub divide the same 10 MHz reference. Detune is applied in the analogue domain, so the beating is imperfect in the way that matters.
STEP 03
Sculpture
A four-pole transistor ladder, hand-biased per unit. At resonance above 17 it self-oscillates into a pure sine you can play as a tenth voice.
STEP 04
Amplification
Discrete VCA, class-A, with a measured 0.02% THD at unity. Drive past that and the curve softens rather than folds — the difference between loud and angry.
STEP 05
Space
Send into the convolution bank, return behind the dry signal by 11 ms. Your ear hears a room before it hears a machine.
05 — Specification
Measured, not claimed
Every unit ships with its own measurement sheet. Hover the cutaway to find where each figure lives.
- Voice architecture
- 8 × discrete analogue, ±0.4 ¢ drift
- Oscillators per voice
- 2 saw-core + sub, hard sync
- Filter
- 4-pole ladder, self-oscillating
- Total harmonic distortion
- 0.02 % @ unity, 20 Hz–20 kHz
- Reverb engine
- Convolution, 40 captured impulses
- Dynamic range
- 124 dB A-weighted
- I/O
- Stereo out, 4 CV in / 4 CV out, MIDI 2.0 TRS
- Memories
- 512 patches × 8 snapshots
- Chassis
- 1.5 mm powder-coated steel, oiled oak cheeks
- Power
- 100–240 V, 34 W idle, toroidal
- Weight / dimensions
- 9.4 kg · 520 × 320 × 96 mm
- Warranty
- 10 years · parts stocked 30 years
06 — Voices
Four factory patches
Pick one and the instrument re-patches itself live — filter, resonance, drive, space, tempo and groove. These are the settings we ship on the bench.
07 — Field notes
From the studios
Pinned to the workshop wall. Drag them around — nobody has tidied them since March.
08 — Editions
Three ways in
Each unit is built to order in Leipzig. Reserve with nothing today; we invoice when your serial enters the queue.
MK-VIII Solo
2 480 €/ unit
- 8-voice analogue engine
- 512 patch memories
- Walnut cheek panels
- 10-year warranty
MK-VIII Studio
3 940 €/ unit
- Everything in Solo
- Matched pair calibration (+0.1 ¢ tolerance)
- Extra CV expansion card
- Bespoke impulse capture of your room
- Two days in the workshop, travel included
Foundry
8 600 €/ pair
- Two serial-adjacent instruments
- Your name etched on the rear plate
- Schematic workshop & repair kit
- Lifetime re-calibration, posted to us
09 — Questions
Reasonable doubts
The five things people ask before they commit four thousand euros to a box of transistors.
Why not just use a plugin?
You should, probably. Plugins recall perfectly, cost nothing and sound closer than ever. The MK-VIII exists because interaction is part of the sound: a knob with resistance changes what you reach for, and six hands on four knobs produces accidents no automation lane writes.
Our own engineers use plugins daily. We simply disagree with the idea that convenience is the only axis worth optimising.
Will it drift out of tune?
Yes, by design — and predictably. All eight voices share a single 10 MHz reference, so relative pitch holds to ±0.4 cent across temperature. What you hear is slow, musical beating between voices, not instability. A full re-calibration takes twenty minutes with two multimeters.
What happens when something fails in twelve years?
Boards are through-hole, schematics are published under CC-BY-SA on day one, and every semiconductor is a current-production part chosen for a thirty-year supply horizon. If we no longer exist, you can still fix it — that was the point of the design brief.
Why only 240 units?
Four people hand-bias and test every voice card. One person can voice eight cards a day, and there are sixty working days in a quarter. That is the entire constraint. We have turned down distribution offers twice.
Is the browser demo the real thing?
No. This page is an honest sketch of the engine — the filter topology, drive curve and impulse reverb are modelled closely, but the hardware adds discrete non-linearity, cross-talk and power-supply sag that no model has captured yet. Consider it the invitation, not the instrument.