dsh-sound-notifier
Sound alerts for DeepSeek Harness (DSH): the Web UI plays a distinct chime when the agent receives a message, starts thinking, calls a tool, or finishes a turn.
Installed as an official DSH bundle (dsh.bundle + dsh.client), so it loads automatically on every DSH start — no per-session activation required.
Features
| Moment | Event | Sound |
|---|---|---|
| Message received | agent/inbox/inserted |
two short pops (660 Hz → 990 Hz) |
| Thinking starts | first reasoning-delta in llm/stream |
rising sweep (392 → 784 Hz, triangle) |
| Tool invoked | tools/pre-execute |
crisp double tick (233 Hz → 311 Hz, triangle) |
| Turn ended | agent/turn-stopping |
major-triad arpeggio (C5–E5–G5) |
- Sounds are synthesized with the Web Audio API — no audio assets.
- Only top-level sessions chime; subagent/workflow activity stays silent.
- Already-played events are not replayed after a page refresh (
sessionStorage).
How it works
The plugin has two halves:
- Host half (
index.js) listens to DSH events, appends them to a bounded queue (256 entries), and serves them over aGET /sound-alerts/events?after=Nendpoint registered through the officialwebServerservice. - Browser half (
client.js) is a standarddsh.clientplugin: the client-modules table injects it intowindow.__DSH_BOOT__, it polls the endpoint every 400 ms, and plays the matching chime.
Requirements
- DeepSeek Harness with the Web UI (
dsh web/dsh --profile web) - pnpm (the
dsh plugincommand forwards to pnpm)
Installation
From the root of your deepseek-harness checkout:
pnpm dsh plugin --profile web add /path/to/dsh-sound-notifier
This links the directory into ~/.dsh/profiles/web, and because package.json declares dsh.bundle, it is appended to the profile's dsh.profile.bundles automatically.
Restart DSH (stop the running pnpm dsh web and start it again) for the plugin to load.
Verify the composed config without restarting:
pnpm dsh --profile web --dump-config # shows a "# == dsh-sound-notifier" layer
Install directly from GitHub (add the printed package key to the profile's pnpm-workspace.yaml allowBuilds on first try, see the official publishing guide):
pnpm dsh plugin --profile web add github:your-name/dsh-sound-notifier
Uninstall
pnpm dsh plugin --profile web remove dsh-sound-notifier
Customizing sounds
Edit the SOUNDS table in client.js:
| Option | Meaning |
|---|---|
freq |
start frequency (Hz) |
endFreq |
sweep target frequency (Hz); omit for no sweep |
dur |
duration (seconds) |
type |
waveform: sine / triangle / square / sawtooth |
gain |
volume (0–1) |
delay |
relative delay (seconds), for sequencing multiple notes |
The local link: install reads the file directly — save and restart DSH.
Project structure
dsh-sound-notifier/
├── package.json # dsh.bundle + dsh.client manifest; exports ./client
├── cordis.patch.yml # plugin row: { id: sound-alerts, name: dsh-sound-notifier }
├── index.js # host half: event listeners, queue, polling endpoint
└── client.js # browser half: dsh.client bundle, polls and plays
Known limitations
- Polling latency: up to ~400 ms between an event and its chime.
- Fixed endpoint:
/sound-alerts/eventsis served by the host half'swebServerroute; without a webserver the events are still recorded but never polled. - Top-level sessions only:
agent.owner === undefinedfiltering means a subagent page opened on its own stays silent. - Internal LLM calls are skipped:
compactionandsession-titlepurposes never trigger the "thinking" chime. - Browser autoplay policy: the audio context is created lazily and
resume()d on each play.
Contributing
Bug reports and pull requests are welcome. Keep changes focused; update this README when behavior changes.
No comments yet. Be the first to write one.