beckon

Distinct sounds for your AI coding agent, so you don't have to watch the terminal.

beckon binds to your agent's lifecycle hooks and plays a different sound depending on what it needs:

A single undifferentiated ding tells you something happened. It doesn't tell you whether to get up.

Install

cargo install beckon-cli   # the binary is called `beckon`
beckon init                # binds the hooks — shows a diff and asks first
beckon test                # hear the active pack

Or from source: git clone https://github.com/Isa1asN/beckon && cd beckon && cargo install --path .

init prints the exact change it will make to ~/.claude/settings.json, copies the file to settings.json.beckon-backup-<timestamp> beside itself, and only then writes. beckon uninstall removes beckon's entries and leaves everything else alone.

It refuses to touch a settings file it cannot round-trip: invalid JSON, invalid UTF-8, a non-object root, or an unfamiliar hooks shape. A file it does accept is re-serialised, so non-canonical formatting (CRLF, tabs, duplicate keys) is normalised. Content is preserved; the backup keeps the original bytes.

Platform support

CI runs the full suite on all three. CI runners have no sound device, so playback there goes through the null backend — the macOS and Windows audio paths compile and are exercised structurally, but nobody has heard them. Reports welcome.

Also built without the embedded-audio feature, which is the shape a fully static musl binary takes: no cpal, no libasound, falling back to a system player.

Requires Claude Code. Other agents are planned; the adapter seam exists and is documented in docs/DESIGN.md.

Commands

beckon test                     # play every sound in the active pack
beckon packs                    # list packs
beckon use cipher               # switch pack
beckon mute 30m                 # quiet for a while (45s, 2h, …); unmute ends it
beckon doctor                   # why is it quiet? every reason, listed
beckon config set volume 0.4
beckon uninstall

When it stays quiet

By default beckon:

• says nothing when a turn finished in under 30 seconds — you were still watching
• always plays a blocking alert, however fast it arrived, because a permission prompt stalls progress
• doesn't repeat the same sound for the same session within 1.5 seconds
• stays quiet for a tool failure, since a failing test is normal work (beckon config set events.tool-failed true to change that)
• ignores a tool you interrupted yourself
• caps concurrent sounds at 8

If it's quiet and you didn't ask it to be, beckon doctor says why.

Several agents at once

Each project gets a stable transposition from a consonant scale, so api-server and worktree-auth sound different with the same pack, and two sounding together harmonise rather than clash. Nothing to configure; disable with beckon config set identity.per_project false.

Rate limiting is scoped per session and per state. A machine-wide throttle would let one agent's completion chime swallow another's permission alert.

Packs

Three ship inside the binary, all original, all CC0:

• aurora — calm starship computer. Soft triangle arpeggios, long reverb.
• cipher — stealth-game alert. Short, dry, cuts through.
• unit-7 — deadpan lab robot. Mechanical FM bleeps, no reverb.

A pack is a TOML file, not a folder of audio. Sounds are synth recipes — oscillators, envelopes, filters — about a kilobyte of text:

[sounds.done]
type = "synth"
reverb = { room = 0.55, mix = 0.34 }

[[sounds.done.layer]]
wave = "triangle"
notes = ["C5", "E5", "G5"]
step_ms = 92
filter = { kind = "lowpass", cutoff_hz = 3200 }

So a pack is provably original, weighs nothing, and can be reviewed as a diff. Auditing a folder of binary blobs for licence provenance is what makes shared sound libraries impractical.

To write one, drop a pack.toml in ~/.local/share/beckon/packs/<id>/. It shadows a built-in of the same name, so you can fork aurora and keep the name. beckon test <id> to hear it.

Your own sounds

You don't need to author a pack to use your own audio:

beckon config set sounds.needs-you ~/sounds/alert.wav

Anything you don't override falls through to the active pack, at every step of the fallback chain — replace failed and rate-limited follows it. wav, ogg, flac and mp3 are supported. The path is checked when you set it, so a typo fails immediately rather than becoming silence you notice days later.

Or write the table directly:

# ~/.config/beckon/config.toml
[sounds]
needs-you = "~/sounds/alert.wav"
done      = "~/sounds/ding.wav"

[sounds] is honoured only in your own config, never in a project's .beckon.toml. A repository you clone can change when beckon makes a noise; it cannot name files on your machine and have them opened by a media decoder.

Sample files are bounded: regular files only, 10 MiB and 30 seconds maximum, and a pack's samples must resolve inside the pack after symlinks are followed.

Footprint

• No daemon. Nothing runs between hooks. Your agent invokes beckon, it decides in ~5ms, hands playback to a detached child, and exits.
• No network at hook time.
• No telemetry.
• Packs are data, never executed.
• Exits 0 unconditionally — bad config, no audio device, corrupt input, panic. beckon binds hooks that block the agent on a non-zero exit, so this is verified against the release binary, where panic = "abort" puts it beyond the reach of cargo test.

Roadmap

• Publish to crates.io
• Hear it on macOS and Windows (CI builds and tests there already)
• beckon install github:user/repo — packs from git
• A browsable community pack index
• SSH: escape sequences so a remote agent alerts your local terminal
• Adapters for Codex, Cursor, Gemini
• npm and Homebrew distribution

Contributing

git clone https://github.com/Isa1asN/beckon && cd beckon
./scripts/install-hooks.sh   # pre-commit: fmt + clippy
./check.sh --release         # everything CI enforces

main requires a pull request and signed commits. If you don't already sign, GitHub will reject the push without much explanation:

git config --global gpg.format ssh
git config --global user.signingkey ~/.ssh/id_ed25519.pub
git config --global commit.gpgsign true

Then add the same key to GitHub under Settings → SSH and GPG keys with key type Signing Key — an authentication key does not count, and that catches most people out.

CI runs fmt, clippy with -D warnings, the suite on Linux/macOS/Windows, a build with no audio backend, an MSRV check, the release-binary safety script, and crates.io packaging. ./check.sh --release covers everything except the other two platforms.

New sound packs are welcome — a pack is a TOML file, so a pull request adding one is reviewable as a diff. Design notes: docs/DESIGN.md.

Licence

Code is MIT OR Apache-2.0. The built-in packs are CC0-1.0.