Hear your coding-agent sessions. Every tool call, file edit, and finished thought becomes a soft, in-key sound — ambient music that moves with Codex, Claude Code, or Gemini CLI.
Claudio listens to lifecycle hook events and turns them into generative, always-in-key ambient music. Twenty tool calls in three seconds become one soft blur of tones, not twenty separate dings — because that's the difference between music and Slack notifications.
It is not a notification system. It's a room your agent session is happening inside. After about twenty minutes you stop hearing it as music and start hearing it as the room — and the room moves with your work.
The
claudio webconsole: each voice is a glowing orb that lights up the instant your agent triggers it. Tune everything live; switch any of 40 presets with a click.
Claude Code users can install the plugin in two lines:
/plugin marketplace add rmtbb/claudio-symphony
/plugin install claudio-symphony
Codex and Gemini CLI are native too. For the universal manual setup, clone the repo and run ./bin/claudio install; it auto-detects all three agents. Or target one explicitly with ./bin/claudio install codex or ./bin/claudio install gemini. See Install for the complete setup.
The hooks wire themselves and your next session starts playing. Each preset renders its sounds the first time it's used, so switching rooms just works. One prerequisite: Python 3 + numpy (pip install numpy); Claudio tries to install numpy for plugin installs, and claudio doctor verifies everything anytime.
macOS / Linux. The plugin hooks call
python3, which is standard there. On Windowspython3often isn't on PATH — use the manual install (or aliaspython3) for now.
- 🎧 It's calm, not noisy. No beeps, no buzzes, no notification DNA. Soothing tones, soft mallets, distant bells — everything tuned to the same key (A = 432 Hz) so it never clashes, no matter how fast your agent is moving.
- 👂 You can hear what your agent is doing. A pluck for a tool call, a chime when a sub-agent returns, a slow bloom when a turn finishes. Without looking, you know whether it is grinding, waiting, or done.
- 🪟 Hear every session at once. Running five agents or terminals? Give each project its own room. Now the chord of your whole workspace is audible — you can tell the quiet one finished and the busy one hit an error, eyes closed.
- 🎹 40 presets, each a complete world. Sunlit music box, lush cathedral, near-silent rainfall, plucked koto, vibraphone lounge, handbell carillon, Rhodes warmth… switch live, mid-session, with one command.
- 🖥️ A console you'll leave open just to watch. A living constellation of your voices blooms in real time as your agent plays them — and the canvas narrates the work: a finished turn breaks across the screen as a slow wave of light, a sub-agent return is a smaller echo, "needs you" pulses the edges. Five physics-driven views (orbs that nudge each other, ripples, ribbons, a particle current, orbiting stars). Click an orb to hear it; drag a slider to tune.
- 🎤 It jams with the room. Hit Listen and Claudio detects what you're humming — or what music you're playing — and re-keys its whole palette to match, live. Put on headphones, enable mic monitor, and your own voice or instrument comes back drenched in Claudio's reverb and delay: you're not listening to your agent anymore, you're in the band with it. Turn on a chord progression and the room cycles changes underneath you while you play.
- 🎚️ Tune everything, or nothing. Sensible defaults out of the box — or open the TUI / web panel and shape every voice, reverb tail, and event mapping to taste.
- 🪶 Featherweight & private. Pure Python, one dependency (numpy), no daemon, no telemetry, no network. Hooks have a 1-second cap and audio playback detaches immediately, so they stay out of the agent's way.
Pick the room you want to work in. whittle is the default — warm wooden taps and celeste chimes, no drone, with a gentle four-chord cycle turning underneath. Switch any time, live, no restart:
claudio preset use cathedralA few of the flagship rooms:
Warm felt taps and knuckle knocks on a wooden desk, answered by celeste chimes and a singing kalimba-mallet over a soft sub-bass. A-major pentatonic — cozy, tactile, never harsh.
The room every fresh install wakes up in — on Claude Code, Codex, and Gemini CLI alike. Dry, in-the-room wood carries the tool calls; bright melodic voices float above it; the bass grounds it without ever turning dark or busy. Out of the box the four-chord pop cycle (A–E–F♯m–D) turns slowly underneath, so a long session quietly becomes a song.
Wooden felt-mallets and kalimba in a sunlit room. A-major pentatonic — universal happy mode, no possible dissonance.
Bright, satisfying, soft attacks. The "thock" of felt on wood for tool calls, kalimba tines for file edits, a small bird-chirp for shell commands, a 9-second major-triad bloom when the agent finishes. Like a music box in a sunny room.
Modal drone bed with a full event palette. Eight voices. Always-on. The Eno / Budd / Frahm lane.
A low drone never stops. Soft felt-piano notes resolve as tools complete; faint airy tones pass by during bash commands; a tuned bell rings when a sub-agent finishes. Nothing has a melody, nothing repeats — the room just moves with the work.
Sparse drops in a still room with rare long blooms. Silence is the canvas.
Most of the time you hear nothing. PostToolUse plays a tiny 250 ms drop — a single bead of water on a still pond. Once a minute or two, a 25-second pad blooms in the distance and recedes. Late-night, deep-focus, almost-silent.
Plucked silk strings and a temple bowl in the A In-Sen scale. Spare, deliberate, like working through a tea ceremony.
A different emotional space from the bright A-major family — minor-leaning, Japanese, unhurried. Koto for melody, bowl for sustains, mokugyo (wooden fish) for accents.
Crystal glass bowls (glassbright), vibraphone (lounge), handbell carillon (peal), steel pan (isleshine), hammered dulcimer (shimmerwire), Rhodes EP (tinewarm), vocal-oo choir (choirloft), nylon guitar (courtyard), ocarina (clayround), santoor (sunraga), and many more — each a complete, render-tested room.
Hear them all in seconds:
claudio auditionwalks every preset on your speakers, or open the web panel and click ▶ on any card.
This is the part you didn't know you wanted. Claudio routes sound per session, with three lookup tiers (most specific wins):
- Session pin —
claudio session pin <id> cathedral→ that one terminal, forever - Directory rule —
claudio rule add "/Users/me/quiet*" rainfall→ every session under that path - Global default —
claudio preset use whittle→ the catch-all
claudio sessions
# # sid pin preset ago src cwd
# 1 33159931 whittle 0m default ~/Projects/shipping-app
# 2 ab12cd34 📌 rainfall 3m pin ~/Projects/quiet-stuffGive your noisy refactor a different voice than your delicate prod fix, and the two coding sessions become audibly distinct. You stop alt-tabbing to check on them — you just hear which one needs you.
claudio web # opens a local, no-deps control panel in your browserA "warm nocturne observatory" you'll actually want to leave open:
- 🌠 A living constellation — each voice is a glowing orb in a slow-drifting galaxy that blooms in real time the instant your agent triggers it — and when one bursts, it pushes its neighbours, so the whole sky wobbles like one fabric. Watch your session play itself.
- 🌊 Five ways to watch — hover the canvas for the ✦ ◍ ≈ ❋ ✸ switcher: the orb constellation (orbs bloom and nudge their neighbours), a still pond of ripples, flowing tides of silk ribbons, a flow field that sprays particles into a drifting current, and gravity where each voice is a star and sounds fling orbiters that slingshot around it — all real-time physics.
- ✨ The light reacts to the work — on top of every view, the moments that matter announce themselves: when your agent finishes a turn a luminous shockwave rolls out across the canvas; a sub-agent returning is a quieter ripple; a notification or permission request glows the edges; session start sweeps, session end fades. Tool calls stay calm, so the big beats land. You can feel the shape of a session from across the room.
- 🎚️ Tune anything live — per-voice gain, reverb, echo; remap any event; switch presets; set per-session rules — all written to the same files the CLI and hooks read, so every change is instant.
- 🔊 The Sounds tab puts cause and effect in one view — "your agent does this → you hear that" up top, the full voice rack right below. Hover any event and hit ✎ to jump straight to that voice's controls; clicking a constellation orb does the same. Three tabs total: Sounds · Music · Setup.
- 🔭 Browse 40 presets in a searchable gallery, audition with one click.
- 🎵 Music tab — global scale override, quantize-to-tempo, MIDI song mode, and a live root-note readout.
- 🎶 Chord progressions — cycle the whole room through the four-chord pop song (I–V–vi–IV), doo-wop, the Andalusian cadence, Pachelbel's Canon, or a lo-fi ii–V–I–vi — or build your own from thirteen chords. The live chord glows in the Music tab and sweeps toward the next change. No daemon, no clock process: the chord is pure wall-time math, so every session and the browser always agree on where you are in the cycle. The drone holds its A pedal underneath — the classic ambient move — or flip Drone follows chords (Setup tab,
claudio drone follow on) and the bed walks the progression's roots with you. And if you've mic-jammed to a new key, the whole progression transposes too. The pop cycle is on out of the box;claudio chords offreturns the room to a static A. CLI:claudio chords pop,claudio chords Am F C G,claudio chords every 12,claudio chords off. - 🎤 Jam with the room — hit Listen and Claudio tunes its key to whatever's around you. (See below.)
- 🎨 Make it yours — the ⋯ → Options menu re-skins the whole console to any accent color (six presets or a custom hue — orbs, glows, sparklines, everything), and sets headphone mode, mic monitoring, and how lively the constellation feels. Saved per-browser.
Pure Python standard library on the backend (127.0.0.1 only), Fraunces + Hanken + Space Mono on the front. No build step, no node_modules.
Hit 🎤 Listen in the top bar and Claudio starts singing along with whatever's happening around your computer. It listens through your mic and detects the note you're humming — or the key of the music you're playing, a guitar, a piano down the hall — and re-keys its entire palette to match, live. Hum a note and watch the button flip to it; put an album on and Claudio slides into its key. The whole room shifts with you; nothing re-renders, so it's instant. Turn the drone bed on (Setup tab, or claudio drone on) and it's the best part: the bed bends to your voice within about half a second — hum a new note and hear the room's foundation lean over to meet you.
Tap to latch it on (it keeps listening until you tap again), or press-and-hold to jam only while held — handy for grabbing one note and letting go. Either way, turning Listen off restores the key you were in before — the mic is a live performance mode, not a permanent setting. The on-button readout shows the note it's settled on and only changes when one note clearly wins, so it reads steady instead of twitchy.
It stays out of its own way without any audio routing or extra dependencies: Claudio is sparse and ambient, so it listens in the gaps between its own notes (ignoring the ~0.4s right after it plays), runs a noise gate so quiet hiss is ignored, and only commits to a note that holds steady for a beat. If it ever does catch its own tail, that just re-affirms the key it's already in — so it settles instead of drifting. Pitch detection is plain Web Audio autocorrelation in the browser; the mic stream is released the moment you toggle it off.
Set the key by hand too: the Root note field in the Music tab (↺ back to A), or
claudio root C/claudio root -2/claudio root offfrom the CLI. Heads-up: an always-on drone preset (e.g.cathedral) gives the mic fewer gaps to listen in — the bright, sparse presets jam most readily.
🎧 Headphone mode — the real jam setup. Wearing headphones? Tell Claudio in ⋯ → Options and Listen gets sharper: with the output in your ears the mic physically can't hear Claudio, so the listening-in-the-gaps filter switches off entirely and it tracks you continuously. Then flip on Mic monitor — mic passthrough with Claudio's reverb and delay on your own voice or instrument, straight into your headphones. You're processed into the same room Claudio is playing in, and now you can actually jam with your Claudio: it follows your key, you ride its texture. A Monitor space slider runs dry ↔ drenched. Monitoring is headphones-only by design, so it can never feed back through your speakers, and none of it leaves the browser.
Take it further: jam over changes. Turn on a chord progression (Music tab → Chord progression, or claudio chords pop) while Listen and the monitor are running — Claudio cycles the four-chord song (or your own changes) underneath you, in your key, while your voice floats through its reverb. A coding-session ambient system is now, quietly, a backing band.
Found a preset + workflow that sounds amazing? Capture it and show it off.
claudio record # record 30 seconds (the default)
claudio record 120 # record up to 5 minutes (300s max)
claudio record --drone # add a faded drone bed for a more song-like clip
claudio record stop # finish early and save right now
claudio record list # see your saved clipsHere's the neat part: because Claudio already knows every sound it plays, recording doesn't touch your mic or system audio — it mixes the exact samples back into one clean track. Only Claudio, no room noise, no other apps, no virtual-audio-device setup. Clips land in recordings/ as a .wav plus a small .m4a for easy sharing.
Optional drone bed. The drone stays off for everyday use, but recordings can opt into one (--drone, or the toggle in claudio web). It lays a low, A-rooted drone under the clip and fades it in and out, so the take feels like one cohesive piece rather than scattered sounds. Off by default — it's there when you want body, gone when you don't.
Prefer buttons? Hit ● Rec in claudio web — pick a length, optionally flip on the drone bed, go work in your sessions, and grab the clip with a built-in player and download link.
🎹 Jukebox (a little easter egg). Don't want to wait for your agent to make sounds? Perform a whole MIDI file through Claudio's voices. Pick the sound kit in the header — any of the 40 presets, independent of the room you're working in. Four are purpose-built song kits with real drums, bass and melodic voices: studio (clean drums · bass · synth), dnb (breakbeats, sub + reese bass, stabs & pad), synthwave (gated-snare 80s with saw bass, lead, arp & poly pad), and lofi (dusty kit, upright bass, Rhodes-y keys & a hazy pad). Each track shows its register, note range, and a density bar, then maps to a sound however you like: to an event (with live fire counts, so the melody can ride whatever your sessions use most), to any voice directly, or ⊞ any sound from any preset. Preview a row with ▶, or hit ✨ smart arrange and Claudio matches drums to percussive sounds and melodies to voices in the right register. Pitch always comes from the MIDI — pitched voices land exactly on every note. The Sounds tab and constellation bloom in time as it plays; start a recording first and you've captured a whole song in your palette; and when a mix sounds right, 💾 save as preset turns it into a self-contained room you can use anywhere. Open it with the . after Claudio (or the Music-tab 🎹 Jukebox button), drop in any .mid, and press play. CLI: claudio play <song> (claudio song import <file.mid> to add one).
🎬 Session mini-tracks — replay your workflow. Every agent session is quietly captured as a tiny symbolic timeline (a few KB of JSON — just what happened, when, no audio). In claudio web, each session in the left rail gets a heavy-traffic sparkline (see the dense, bursty stretches at a glance) and a ▶ replay button. Replay re-runs your actual workflow as music through any preset — so you can audition the same session across presets and tune until it sounds great. Idle gaps are compressed so the busy parts stay punchy. Then ● render WAV bounces it to a shareable clip, or ⤓ score saves the tiny .score.json (far lighter to share than a WAV, and replayable anywhere). From the CLI: claudio replay list, claudio replay <session|latest> [--preset X] [--tempo 1] [--loop] [--render].
Share it! 🎧 Post your clip with the preset name so others can hear how their sessions could sound. The more sounds people share, the richer Claudio gets — and if you tune a combination you love, open a PR with the preset too.
Paste these into Claude Code:
/plugin marketplace add rmtbb/claudio-symphony
/plugin install claudio-symphony
That's it — the hooks wire themselves, and on your next session Claudio renders the default room and starts playing. The claudio command is on your PATH (claudio web opens the console). One prerequisite: Python 3 with numpy (pip install numpy); if it's missing, Claudio drops a one-line note in logs/SETUP_NEEDED.txt telling you exactly that.
Claudio ships a native .codex-plugin/plugin.json and a Codex-compatible lifecycle hook bundle. Install it from the CLI (or add the marketplace and choose Claudio Symphony from the Plugins directory in the Codex app):
codex plugin marketplace add rmtbb/claudio-symphony
codex plugin add claudio-symphony@claudioOn first use, review and trust the Claudio hook definitions when Codex prompts you (or open /hooks in the Codex CLI). Codex deliberately requires trust again whenever a hook definition changes.
Prefer to drive it by hand, use Gemini CLI, or avoid plugins? Use the universal manual install below. Don't install both plugin and manual hooks for the same agent, or each event will play twice.
git clone https://github.com/rmtbb/claudio-symphony.git
cd claudio-symphony
python3 install.py # deps + render samples + write starter config
./bin/claudio install # auto-detect Claude Code, Codex, and Gemini CLITarget agents explicitly when you want tighter control:
./bin/claudio install codex
./bin/claudio install claude gemini
./bin/claudio install all
./bin/claudio uninstall codexThe installer preserves unrelated hooks and backs up every settings file it changes. It writes Claude Code hooks to ~/.claude/settings.json, Codex hooks to ~/.codex/hooks.json, and Gemini CLI hooks to ~/.gemini/settings.json. Start a new agent session and listen. For Codex, open /hooks once to trust the new definitions.
Gemini's BeforeTool, AfterTool, BeforeAgent, AfterAgent, and PreCompress events are normalized to the same musical vocabulary as the other agents, so presets remain portable. Other command-hook systems can use the generic JSON bridge.
👉 There's a full UI. Run
claudio webto open the browser control panel — a live constellation of your voices where you can tune everything, browse and switch all 40 presets, build your own, and record clips. No build step, runs locally on127.0.0.1. See The web control panel.
Requirements
- 🖥️ macOS, Linux, or Windows. Claudio auto-detects the best audio player on your system (override with
CLAUDIO_PLAYER):- 🍎 macOS — built-in
afplay. Nothing to install; full quality. - 🐧 Linux —
ffplay(fromffmpeg) for full quality; falls back tompv/sox/pw-play/paplay/aplay.sudo apt install ffmpeg(or your distro's equivalent) recommended. - 🪟 Windows —
ffplay(winget install -e --id Gyan.FFmpeg) for full quality; falls back to PowerShell MediaPlayer (volume only) thenwinsound(basic blips).
- 🍎 macOS — built-in
- 🐍 Python 3.9+ with numpy — still the only Python dependency. (Players are auto-detected, never auto-installed.)
- 💾 ~250 MB free for generated samples (one-time render, then static)
Song-mode pitch micro-tuning is native on afplay/ffplay/mpv/sox; on volume-only backends (pw-play/paplay/Windows MediaPlayer) it's reproduced by an offline numpy pre-render (cached under state/rate_cache/), so it still works everywhere. The curses tuner (claudio tune) needs pip install windows-curses on Windows — or just use claudio web, which works everywhere.
Tip: add alias claudio='~/path/to/claudio-symphony/bin/claudio' to your shell profile (macOS/Linux). On Windows, run via bin\claudio.cmd.
claudio status # what's installed and active
claudio preset list # see all 40 presets
claudio preset use cathedral # switch live (no agent restart)
claudio off / claudio on # silence everything / restore
claudio web # the browser control panel
claudio tune # interactive terminal tuner (TUI)
claudio audition # hear every preset, then pick one
claudio test # walk every voice in the active preset
claudio volume 0.4 # master gain
claudio coffee # ☕ support the project (see below)─────────────── claudio tune ─────────────────────────────────
preset: meadow master: ▰▰▰▰▰▱▱▱▱▱ 0.55 ✓ saved
─ Voices — [gain] press t for mioi ──────────────────────────
▶ mallet gain ▰▰▰▰▰▱▱▱ 0.55 mioi ▰▰▱▱ 0.35s
kalimba gain ▰▰▰▰▰▱▱▱ 0.50 mioi ▰▰▱▱ 0.30s
bloom gain ▰▰▰▰▰▱▱▱ 0.55 mioi ▰▰▰▰ 10.00s
─ Events ─────────────────────────────────────────────────────
PreToolUse → wood
· Bash → bird
PostToolUse → mallet
· Edit → kalimba
· on_failure → chime
Stop → bloom
─ TAB pane │ ↑↓ row │ ←→ value │ SPC play │ s save │ q ──────
Arrow keys move sliders, SPACE previews, s saves, q saves + quits. Changes are live — the next event picks them up without restarting your agent.
claudio voice mallet gain 0.45 # per-voice gain in the active preset
claudio voice mallet mioi 0.5 # rate-limit (min seconds between hits)
claudio map PostToolUse:Edit pluck # remap by tool
claudio map PostToolUse:on_failure - # silence the failure variant
claudio scale use A_lydian # global scale override
claudio root C # live-transpose the whole room (root off → back to A)
claudio chords pop # cycle the four-chord song (chords Am F C G · every 12 · off)
claudio song use mario # drive a voice from a MIDI melodyThe easy way — build one in the browser. Open claudio web, hit ✚ Build a preset, and you get a palette of every voice from all 40 presets. Audition any sound with ▶, pick the ones you love, name it, and hit Create — Claudio copies those samples into a new, self-contained preset that works everywhere instantly (start blank, or duplicate an existing preset and add to it). No code, no render step.
- Swap a single sound — in the Sounds tab, the ↺ button on any voice lets you replace just that one sound with any voice from any preset, keeping its level, echo, and event mappings.
- Manage your presets — custom presets you build get rename (✎) and delete (🗑) right on their gallery card; the 40 shipped presets are protected and can't be deleted.
The hands-on way — author it in code. Each preset is three things:
presets/<your-preset>/
├── preset.json # voice configs + event → voice mapping
├── render.py # generates the WAV samples
└── samples/ # output (gitignored — generated by render.py)
render.py imports the shared DSP helpers from the top-level synth.py (FFT-convolved reverb, ADSR envelopes, FFT lowpass, A = 432 frequency math) and writes WAVs into samples/<voice>/.
There's a full composer's brief covering scale choices, palette design, and the anti-machine-gun strategies (per-voice MIOI, pressure accumulators, voice stealing, reverb-as-glue). If you build a room you love, open a PR — we'd genuinely love to hear it.
Codex / Claude Code / Gemini CLI event
│ command hook fires (1s cap; Gemini names are normalized)
▼
event.py (~50 ms total)
│ ├─ reads the event JSON from stdin
│ ├─ resolves the preset: session pin → cwd rule → global default
│ ├─ maps event → voice via preset.json
│ ├─ rate-limits per voice (drops too-soon hits, banks "pressure"
│ │ that bumps the next note's amplitude)
│ ├─ picks a pitch melodically (Markov-weighted, stays in scale)
│ └─ launches the detected player detached (afplay / ffplay / … via `audio.py`)
▼
(your speakers)
event.py is stateless except for tiny per-voice timestamp files. drone.py is the only long-running process, and it auto-exits after 10 minutes of silence. No daemon to babysit, no port to conflict, no config server. Pure Python, one dependency.
Three principles shaped every design choice:
- Inverse frequency → prominence. Rare events get the foreground; common events whisper. The end-of-turn bloom feels loud because it happens once every few minutes; tool-starts get the softest, most disposable voice because they fire 5–30× a minute.
- No tempo, no melody, no notification DNA. With nothing to predict and no memorable hook, the ear settles into landscape mode instead of flinch-mode. The instant something sounds like a notification, your nervous system flags every event as a possible interruption.
- The reverb tail is the glue. A long tail on every voice means dense bursts smear into a wash rather than stack as discrete hits. The same patch handles the calm session and the chaotic one — the wash gets thicker, not louder.
Read the full composer's brief — it's worth your time even if you never touch the code.
I wanted my coding-agent sessions to make soothing, pleasant sounds — all in key, all cohesive — so that AI-assisted coding simply felt nicer to sit inside.
Then something happened while I was building it: I started to rely on it. I could suddenly hear what was happening across all of my active coding sessions — which one was busy, which had gone quiet, which just finished. What began as a small idea turned into something I genuinely don't want to use a coding agent without.
Slack pings condition you to flinch. A cathedral does not. A bell tuned to the room you're working in becomes a fact about the room, not an interruption. The hope is that the room starts to move with your work — and that you fall for it the way I did. 🌙
Claudio is free and MIT-licensed, built and maintained in my spare time. If it makes your sessions a little nicer, an on-chain tip means the world and keeps new presets coming. No platform, no cut, no signup — just send to whichever chain you already use:
| Chain | Address | Also accepts | Scan |
|---|---|---|---|
| Bitcoin (BTC) | 3Ht6H531DyWjgsxV289cWLYKJG8tVza2P5 |
— |  |
| Ethereum (ETH) | 0xB239b34e42b4Bc571158e48a779099950C81C7d8 |
USDC (ERC-20) · any EVM token |  |
| Solana (SOL) | 2PiDya4hpbpki7iUceyE1uYBDR3dQgW6uCN27JVJUHwS |
USDC (SPL) |  |
| Dogecoin (DOGE) | D5HMGHjcRnvdcNY5LGAb1EcMpBm8FD8YHN |
— |  |
The Ethereum and Solana addresses receive their native coin first (ETH / SOL) and also take USDC on that chain — same address, no extra step. Send tokens only on the network listed above so they land somewhere I can reach them. You can also pull these up any time with claudio coffee, or via ⋯ → Buy me a coffee in claudio web (with QR + one-tap copy).
Verifying these addresses: all four live in
donate.jsonas the single source of truth, and the app reads the same file. If an address here ever disagrees withdonate.jsonor your wallet, treat it as a tampered diff and don't send. (These are public receiving addresses — never a private key.)
⭐ Can't tip? Starring the repo and sharing it helps just as much.
- The two design briefs in
docs/—SONIC_FRAMEWORK.mdandTRIGGER_SURFACE.md— were drafted as the spec for this project, and are worth reading on their own. - DSP synthesis: pure numpy. Synthetic-IR FFT convolution for the reverb. No scipy, no external audio libraries.
- Inspired by Brian Eno (Music for Airports), Harold Budd (The Pearl), Stars of the Lid, Nils Frahm (Felt), William Basinski, Susumu Yokota.
- Tuning: A = 432 Hz throughout. Drone fundamentals are strictly just-intoned (3:2 root-to-fifth); upper voices are equal-tempered.
MIT. Do whatever you want with it. If you build a preset you love, we'd love to see it. 💫
tune the drone first
get the reverb right
add the pluck
everything else is decoration on those three foundations