This information is current as of 2017-03-21 but is subject to rapid change.

The input data stream to a networked plugin consists of a sequence of blocks, each corresponding to a “chunk” of stereo audio samples (e.g. between 500 and 1000). Each block is comprised of a header followed by a chunk of input data (PCM audio for effects, MIDI event data for synthesizers). Expressed in C:

#pragma pack(push, 1)
 
typedef struct
{
    UINT16 frameCount;   // defines CHUNKSIZE below
    UINT16 midiCount;    // will be 0 for a filter plugin
    UINT32 timeStamp;    // start time of this block, in units of sampling period, since song start
} SendDataHeader;
 
typedef struct
{
    UINT8 status;
    UINT8 channel;
    UINT8 data1;
    UINT8 data2;
    UINT32 startFrame;   // start time of this event, in units of sampling period, since block start
} MIDIMessageInfoStruct;
 
typedef struct
{
    UINT16  mainByteCount;      // or total bytecount if uncompressed
    UINT16  corrByteCount;      // 0 = lossy compression, 0xFFFF = uncompressed
} SampleDataHeader;
 
// Uncompressed sample data are represented like this:
typedef struct
{
    FLOAT left[CHUNKSIZE];  // Uncompressed PCM audio data, stereo, non-interleaved,
    FLOAT left[CHUNKSIZE];  // normalized to range [-1.0f, 1.0f]
} UncompressedStereoSampleData;
 
// Compressed sample data are represented like this (not syntactically correct C):
typedef struct
{
    BYTE lossyData[mainByteCount];
    BYTE losslessCorrectionData[corrByteCount]; // zero length for lossy-compressed data
} CompressedStereoSampleData;
 
#pragma pack(pop)

For a synthesizer plugin, the header midiCount will be non-zero, and the header is followed by that many MidiMessageInfoStruct items. For a filter plugin, midiCount will be zero.

The output stream consists of a SampleDataHeader followed by EITHER UncompressedStereoSampleData OR CompressedStereoSampleData.

The WavPack audio compression library is used, which can produce either one block of lossy-compressed data, and optionally a second block of “correction” data which can be combined with the first block to reconstruct the original samples exactly (lossless compression).

The header structure will be expanded to allow:

• sending MIDI data to filters (for e.g. parameter automation), in addition to audio
• sending audio to synths (e.g. sidechain input to a vocoder), in addition to MIDI
• incorporating audio compression algorithms other than WavPack
• adding remote parameter-change data, to facilitate remote GUIs and parameter automation/modulation with greater precision than is supported by MIDI.