diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..3e2f2a8
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,2 @@
+*.lock
+target
diff --git a/Cargo.toml b/Cargo.toml
new file mode 100644
index 0000000..9ed6213
--- /dev/null
+++ b/Cargo.toml
@@ -0,0 +1,30 @@
+[package]
+name = "djdeck"
+version = "0.1.0"
+edition = "2024"
+
+[[bin]]
+name = "djdeck"
+path = "src/main.rs"
+
+[dependencies]
+ratatui = "0.30"
+crossterm = "0.29"
+pipewire = "0.9.2"
+symphonia = { version = "0.5", features = ["mp3", "flac", "ogg", "wav", "aac"] }
+rosc = "0.10"
+tokio = { version = "1", features = ["full"] }
+crossbeam-channel = "0.5"
+serde = { version = "1", features = ["derive"] }
+serde_json = "1"
+anyhow = "1"
+dirs = "5"
+walkdir = "2"
+parking_lot = "0.12"
+tracing = "0.1"
+tracing-subscriber = { version = "0.3", features = ["env-filter"] }
+
+[profile.release]
+opt-level = 3
+lto = true
+codegen-units = 1
diff --git a/README.md b/README.md
index f5e94ef..7b20334 100644
--- a/README.md
+++ b/README.md
@@ -60,7 +60,7 @@ analyseplugin ladspa-rubberband.so
 ### Rust
 
 ```
-pipewire-rs  (pipewire = "0.8")
+pipewire-rs  (pipewire = "0.9.2")
 symphonia    (audio decoding)
 ratatui + crossterm
 rosc         (OSC)
diff --git a/src/audio.rs b/src/audio.rs
new file mode 100644
index 0000000..ee1dcf0
--- /dev/null
+++ b/src/audio.rs
@@ -0,0 +1,417 @@
+/// Audio engine
+///
+/// Responsibilities:
+///   - Decode audio files (symphonia) into interleaved stereo f32 PCM
+///   - Drive one PipeWire playback stream per deck
+///   - Each stream targets the deck's filter-chain sink node
+///     (djdeck.chain.<x> input side) so all DSP happens in the graph
+///   - Report position, VU, and load status back to the app via channel
+///
+/// No DSP happens here. Pitch/tempo/EQ all live in the PipeWire graph.
+use std::{
+    path::PathBuf,
+    sync::{Arc, Mutex},
+    thread,
+};
+
+use anyhow::{anyhow, Result};
+use crossbeam_channel::{bounded, Receiver, Sender};
+use symphonia::core::{
+    audio::{AudioBufferRef, Signal},
+    codecs::DecoderOptions,
+    formats::FormatOptions,
+    io::MediaSourceStream,
+    meta::MetadataOptions,
+    probe::Hint,
+};
+
+use crate::{deck::DeckId, effect_chain::EffectChain};
+
+// ─── Commands ───────────────────────────────────────────────────────────────
+
+#[derive(Debug)]
+pub enum AudioCmd {
+    Load { deck: DeckId, path: PathBuf },
+    Play(DeckId),
+    Pause(DeckId),
+    Stop(DeckId),
+    Seek { deck: DeckId, position: f32 },
+    SetCue(DeckId),
+    GotoCue(DeckId),
+    // Volume is handled by wpctl; we just record it in state.
+    // Fader changes are applied directly from app.rs via EffectChain::set_volume.
+}
+
+// ─── Status updates ─────────────────────────────────────────────────────────
+
+#[derive(Debug, Clone)]
+pub enum AudioStatus {
+    Loaded {
+        deck: DeckId,
+        duration_samples: u64,
+        sample_rate: u32,
+        name: String,
+        /// PipeWire object ID of the stream node (for wpctl volume)
+        stream_node_id: u32,
+    },
+    Position { deck: DeckId, position: u64 },
+    Vu { deck: DeckId, l: f32, r: f32 },
+    TrackEnded(DeckId),
+    Error { deck: DeckId, msg: String },
+}
+
+// ─── Per-deck internal state ──────────────────────────────────────────────────
+
+struct DeckState {
+    id: DeckId,
+    playing: bool,
+    pcm: Vec<f32>,   // interleaved stereo f32
+    sample_rate: u32,
+    read_pos: usize, // byte index into pcm (always even)
+    cue_pos: usize,
+}
+
+impl DeckState {
+    fn new(id: DeckId) -> Self {
+        Self {
+            id,
+            playing: false,
+            pcm: Vec::new(),
+            sample_rate: 44100,
+            read_pos: 0,
+            cue_pos: 0,
+        }
+    }
+}
+
+// ─── Public handle ──────────────────────────────────────────────────────────
+
+pub struct AudioEngine {
+    pub cmd_tx: Sender<AudioCmd>,
+    pub status_rx: Receiver<AudioStatus>,
+}
+
+impl AudioEngine {
+    pub fn new() -> Result<Self> {
+        let (cmd_tx, cmd_rx) = bounded::<AudioCmd>(64);
+        let (status_tx, status_rx) = bounded::<AudioStatus>(512);
+
+        thread::Builder::new()
+            .name("audio-engine".into())
+            .spawn(move || {
+                if let Err(e) = engine_thread(cmd_rx, status_tx) {
+                    tracing::error!("Audio engine crashed: {e}");
+                }
+            })?;
+
+        Ok(Self { cmd_tx, status_rx })
+    }
+}
+
+// ─── Engine thread ─────────────────────────────────────────────────────────
+
+fn engine_thread(cmd_rx: Receiver<AudioCmd>, status_tx: Sender<AudioStatus>) -> Result<()> {
+    use pipewire::{context::ContextRc, main_loop::MainLoopRc, stream::StreamRc};
+
+    let main_loop = MainLoopRc::new(None)?;
+    let context = ContextRc::new(&main_loop, None)?;
+    let core = context.connect_rc(None)?;
+
+    // Shared per-deck state (written by command thread, read by PW callback)
+    let deck_states: Vec<Arc<Mutex<DeckState>>> = (0..4)
+        .map(|i| Arc::new(Mutex::new(DeckState::new(DeckId::from_index(i).unwrap()))))
+        .collect();
+
+    // Build one PipeWire stream per deck
+    let mut _streams: Vec<StreamRc> = Vec::new();
+    let mut _listeners = Vec::new();
+
+    for (i, ds) in deck_states.iter().enumerate() {
+        let deck_id = DeckId::from_index(i).unwrap();
+        let ds_cb = Arc::clone(ds);
+        let status_tx_cb = status_tx.clone();
+
+        // Target: the input (sink) side of this deck's filter-chain node
+        let target_name = format!("input.{}", EffectChain::node_name_for(deck_id));
+
+        let props = pipewire::properties::properties! {
+            *pipewire::keys::MEDIA_TYPE     => "Audio",
+            *pipewire::keys::MEDIA_CATEGORY => "Playback",
+            *pipewire::keys::MEDIA_ROLE     => "Music",
+            *pipewire::keys::APP_NAME       => "djdeck",
+            *pipewire::keys::NODE_NAME      => format!("djdeck.deck.{}", deck_id.node_suffix()),
+            // Route to our filter-chain sink
+            "node.target"                   => target_name,
+        };
+
+        let stream = StreamRc::new(
+            core.clone(),
+            &format!("djdeck-deck-{}", deck_id.label()),
+            props,
+        )?;
+
+        let listener = stream
+            .add_local_listener_with_user_data(())
+            .process(move |stream: &pipewire::stream::Stream, _| {
+                let mut ds_guard = ds_cb.lock();
+                let ds = ds_guard.as_mut().unwrap();
+                if let Some(mut buf) = stream.dequeue_buffer() {
+                    let datas = buf.datas_mut();
+                    if datas.is_empty() { return; }
+
+                    let data = &mut datas[0];
+                    let chunk = data.chunk_mut();
+                    let n_frames = (chunk.size() / 8) as usize; // f32 stereo = 8 bytes
+                    let out_bytes = data.data().unwrap();
+                    let out = unsafe {
+                        std::slice::from_raw_parts_mut(
+                            out_bytes.as_ptr() as *mut f32,
+                            n_frames * 2,
+                        )
+                    };
+
+                    let mut vu_l = 0.0f32;
+                    let mut vu_r = 0.0f32;
+
+                    if ds.playing && !ds.pcm.is_empty() {
+                        for frame in 0..n_frames {
+                            if ds.read_pos + 1 >= ds.pcm.len() {
+                                // End of track
+                                ds.playing = false;
+                                out[frame * 2]     = 0.0;
+                                out[frame * 2 + 1] = 0.0;
+                                let _ = status_tx_cb.try_send(AudioStatus::TrackEnded(ds.id));
+                                continue;
+                            }
+                            let l = ds.pcm[ds.read_pos];
+                            let r = ds.pcm[ds.read_pos + 1];
+                            ds.read_pos += 2;
+                            out[frame * 2]     = l;
+                            out[frame * 2 + 1] = r;
+                            vu_l = vu_l.max(l.abs());
+                            vu_r = vu_r.max(r.abs());
+                        }
+                    } else {
+                        out.fill(0.0);
+                    }
+
+                    *chunk.offset_mut() = 0;
+                    *chunk.stride_mut() = 8;
+                    *chunk.size_mut()   = (n_frames * 8) as u32;
+
+                    let pos = (ds.read_pos / 2) as u64;
+                    let _ = status_tx_cb.try_send(AudioStatus::Position { deck: ds.id, position: pos });
+                    let _ = status_tx_cb.try_send(AudioStatus::Vu { deck: ds.id, l: vu_l, r: vu_r });
+                }
+            })
+            .register()?;
+
+        // Connect the stream with F32LE stereo format
+        let pod_bytes = make_f32_stereo_pod()?;
+        let mut params = [pipewire::spa::pod::Pod::from_bytes(&pod_bytes).unwrap()];
+        stream.connect(
+            pipewire::spa::utils::Direction::Output,
+            None,
+            pipewire::stream::StreamFlags::AUTOCONNECT
+                | pipewire::stream::StreamFlags::MAP_BUFFERS
+                | pipewire::stream::StreamFlags::RT_PROCESS,
+            &mut params,
+        )?;
+
+        _listeners.push(listener);
+        _streams.push(stream);
+    }
+
+    // Command processing thread — decodes files and mutates deck states
+    let ds_cmd = deck_states.clone();
+    let st_cmd = status_tx.clone();
+    thread::spawn(move || {
+        for cmd in cmd_rx.iter() {
+            handle_cmd(cmd, &ds_cmd, &st_cmd);
+        }
+    });
+
+    main_loop.run();
+    Ok(())
+}
+
+fn handle_cmd(
+    cmd: AudioCmd,
+    states: &[Arc<Mutex<DeckState>>],
+    status_tx: &Sender<AudioStatus>,
+) {
+    match cmd {
+        AudioCmd::Load { deck, path } => {
+            let name = path.file_name()
+                .and_then(|n| n.to_str())
+                .unwrap_or("unknown")
+                .to_string();
+
+            match decode_file(&path) {
+                Ok((pcm, sr)) => {
+                    let total = (pcm.len() / 2) as u64;
+                    {
+                        let mut ds = states[deck as usize].lock().unwrap();
+                        ds.pcm = pcm;
+                        ds.sample_rate = sr;
+                        ds.read_pos = 0;
+                        ds.cue_pos = 0;
+                        ds.playing = false;
+                    }
+                    // stream_node_id: we don't have it from here — the app queries
+                    // pw-dump after connecting. Emit 0 as placeholder; app fills it in.
+                    let _ = status_tx.send(AudioStatus::Loaded {
+                        deck,
+                        duration_samples: total,
+                        sample_rate: sr,
+                        name,
+                        stream_node_id: 0,
+                    });
+                }
+                Err(e) => {
+                    let _ = status_tx.send(AudioStatus::Error { deck, msg: e.to_string() });
+                }
+            }
+        }
+        AudioCmd::Play(deck) => {
+            let mut ds = states[deck as usize].lock().unwrap();
+            if !ds.pcm.is_empty() {
+                ds.playing = true;
+            }
+        }
+        AudioCmd::Pause(deck) => {
+            let mut ds = states[deck as usize].lock().unwrap();
+            ds.playing = false;
+        }
+        AudioCmd::Stop(deck) => {
+            let mut ds = states[deck as usize].lock().unwrap();
+            ds.playing = false;
+            ds.read_pos = 0;
+        }
+        AudioCmd::Seek { deck, position } => {
+            let mut ds = states[deck as usize].lock().unwrap();
+            let total = ds.pcm.len();
+            ds.read_pos = ((total as f32 * position.clamp(0.0, 1.0)) as usize / 2) * 2;
+        }
+        AudioCmd::SetCue(deck) => {
+            let mut ds = states[deck as usize].lock().unwrap();
+            ds.cue_pos = ds.read_pos;
+        }
+        AudioCmd::GotoCue(deck) => {
+            let mut ds = states[deck as usize].lock().unwrap();
+            ds.read_pos = ds.cue_pos;
+        }
+    }
+}
+
+// ─── Symphonia decode ─────────────────────────────────────────────────────────────
+
+/// Decode any supported audio file into interleaved stereo f32 PCM.
+fn decode_file(path: &PathBuf) -> Result<(Vec<f32>, u32)> {
+    let file = std::fs::File::open(path)?;
+    let mss = MediaSourceStream::new(Box::new(file), Default::default());
+
+    let mut hint = Hint::new();
+    if let Some(ext) = path.extension().and_then(|e| e.to_str()) {
+        hint.with_extension(ext);
+    }
+
+    let probed = symphonia::default::get_probe().format(
+        &hint,
+        mss,
+        &FormatOptions::default(),
+        &MetadataOptions::default(),
+    )?;
+
+    let mut format = probed.format;
+    let track = format
+        .tracks()
+        .iter()
+        .find(|t| t.codec_params.codec != symphonia::core::codecs::CODEC_TYPE_NULL)
+        .ok_or_else(|| anyhow!("No audio track found"))?
+        .clone();
+
+    let sr = track.codec_params.sample_rate.unwrap_or(44100);
+    let channels = track.codec_params.channels.map(|c| c.count()).unwrap_or(2);
+    let mut decoder = symphonia::default::get_codecs()
+        .make(&track.codec_params, &DecoderOptions::default())?;
+
+    let track_id = track.id;
+    let mut pcm: Vec<f32> = Vec::new();
+
+    loop {
+        let packet = match format.next_packet() {
+            Ok(p) => p,
+            Err(symphonia::core::errors::Error::IoError(_)) => break,
+            Err(symphonia::core::errors::Error::ResetRequired) => break,
+            Err(e) => return Err(e.into()),
+        };
+        if packet.track_id() != track_id { continue; }
+
+        let decoded = decoder.decode(&packet)?;
+        push_samples(&mut pcm, decoded, channels);
+    }
+
+    Ok((pcm, sr))
+}
+
+fn push_samples(out: &mut Vec<f32>, buf: AudioBufferRef, channels: usize) {
+    macro_rules! push {
+        ($buf:expr, $scale:expr) => {{
+            let frames = $buf.frames();
+            for f in 0..frames {
+                let l = *$buf.chan(0).get(f).unwrap_or(&Default::default());
+                let r = if channels >= 2 {
+                    *$buf.chan(1).get(f).unwrap_or(&Default::default())
+                } else {
+                    l
+                };
+                out.push(l as f32 * $scale);
+                out.push(r as f32 * $scale);
+            }
+        }};
+    }
+
+    match buf {
+        AudioBufferRef::F32(b)  => push!(b, 1.0_f32),
+        AudioBufferRef::F64(b)  => push!(b, 1.0_f32),
+        AudioBufferRef::S16(b)  => push!(b, 1.0 / 32768.0_f32),
+        AudioBufferRef::S32(b)  => push!(b, 1.0 / 2147483648.0_f32),
+        AudioBufferRef::U8(b)   => push!(b, 1.0 / 128.0_f32),
+        _ => {}
+    }
+}
+
+// ─── PipeWire format pod helper ───────────────────────────────────────────────
+
+fn make_f32_stereo_pod() -> Result<Vec<u8>> {
+    use pipewire::spa::pod::serialize::PodSerializer;
+    use pipewire::spa::pod::{object, property, Value};
+    use pipewire::spa::param::audio::AudioFormat;
+    use pipewire::spa::param::format::{FormatProperties, MediaSubtype, MediaType};
+    use pipewire::spa::param::ParamType;
+    use pipewire::spa::utils::SpaTypes;
+
+    let bytes = PodSerializer::serialize(
+        std::io::Cursor::new(Vec::new()),
+        &Value::Object(object!(
+            SpaTypes::ObjectParamFormat,
+            ParamType::EnumFormat,
+
+    let bytes = PodSerializer::serialize(
+        std::io::Cursor::new(Vec::new()),
+        &Value::Object(object!(
+            SpaTypes::ObjectParamFormat,
+            ParamType::EnumFormat,
+            property!(FormatProperties::MediaType,    Id, MediaType::Audio),
+            property!(FormatProperties::MediaSubtype, Id, MediaSubtype::Raw),
+            property!(FormatProperties::AudioFormat,   Id,  AudioFormat::F32LE),
+            property!(FormatProperties::AudioRate,     Int, 44100i32),
+            property!(FormatProperties::AudioChannels, Int, 2i32),
+        )),
+    )?
+    .0
+    .into_inner();
+
+    Ok(bytes)
+}
\ No newline at end of file