Resound

Portable memory for Discord conversations. Resound records voice locally, transcribes with the provider you choose, and writes durable Markdown, JSONL, VTT, and SRT artifacts that you own.

No hosted service, web dashboard, or database is required. Strata and TotalRecall are optional destinations, not dependencies.

pnpm cli mock "Engineering Standup"
pnpm cli validate <session>
pnpm cli export <session> --format md

Contents

• Why Resound · Quick start · Real recording
• What you can do · How it works · Discord bot
• Project status · Documentation · License

---

Why Resound

• Local-first — capture runs on the operator's machine and session files stay under operator control.
• Vendor-neutral — use local Whisper or any OpenAI-compatible transcription endpoint.
• Portable — every session is a self-describing folder, not a row trapped in a service database.
• Consent-aware — recording is announced and consent events are preserved in the session manifest.
• Composable — keep the files, post them to a webhook, or send them to optional Strata and TotalRecall sinks.

Quick start

Prerequisites: Node.js 20 or newer and pnpm 9.15.0.

nvm use
corepack enable
pnpm install
pnpm build
pnpm test

# Complete offline smoke test: record, transcribe, and export mock data.
pnpm cli mock "Engineering Standup"

Run the CLI from source with pnpm cli <command>. After pnpm build, the compiled entry point is apps/cli/dist/index.js.

Real recording

On macOS, route call audio through a virtual input such as BlackHole, then let Resound capture it alongside your microphone:

pnpm cli doctor --mode local-capture
pnpm cli audio devices
pnpm cli record \
  --title "Client Call" \
  --system <blackhole-index> \
  --mic <microphone-index> \
  --participants "Robert,Client"

To transcribe an existing recording instead:

pnpm cli transcribe ./meeting.m4a \
  --title "Q3 Planning" \
  --provider local-whisper \
  --participants "Robert,Ashley,Jelena"

See the recording guide for macOS audio routing and the provider guide for local Whisper and OpenAI-compatible configuration.

What you can do

Command	What it does
resound doctor	Check recorder and transcriber readiness
resound audio devices	List macOS audio input devices
resound record	Capture system audio and microphone, then transcribe
resound transcribe <file>	Turn an existing audio file into a session
resound mock <title>	Create a complete offline test session
resound sessions list	List local sessions
resound sessions show <session>	Inspect a session manifest and transcript preview
resound validate <session>	Validate consent, manifest, outputs, and Kujo checks
resound export <session> --format all	Write Markdown, JSONL, VTT, and SRT outputs
resound summarize <session>	Regenerate the Markdown summary
resound action-items <session>	Regenerate extracted action items
resound sink <target> <session>	Send artifacts to stdout, a folder, webhook, Strata, or TotalRecall

Run pnpm cli --help for the complete command list and pnpm cli <command> --help for every flag.

How it works

audio source
    │
    ├─ local system audio + microphone
    ├─ existing audio file
    └─ Discord-native receiver (experimental)
    │
    ▼
transcriber ── local Whisper or OpenAI-compatible API
    │
    ▼
portable session folder
    ├─ manifest.json
    ├─ transcript.jsonl
    ├─ transcript.md
    ├─ transcript.vtt / transcript.srt
    ├─ summary.md
    └─ action-items.md

Discord bot

The bot provides consent, status, and recording controls through /resound. It supports four recorder modes:

Mode	Purpose
mock	Exercise the complete workflow without real audio
local-capture	Control system-audio and microphone capture on the bot host
discord-native	Receive Discord voice through the Pycord sidecar
auto	Prefer Discord-native and fall back when local capture is configured

cp .env.example .env
pnpm bot:register
pnpm bot:start

Discord-native voice receive is still experimental because Discord's DAVE/E2EE behavior requires live acceptance testing across real calls. For reliable production use today, prefer local capture or transcribe an existing recording. The provider guide documents the constraint and current receiver status.

Repository layout

apps/cli/             command-line interface
apps/bot/             Discord slash-command bot
packages/audio/       capture and Discord receiver adapters
packages/core/        sessions, manifests, consent, validation, storage
packages/transcribers provider adapters
packages/exporters/   Markdown, JSONL, VTT, SRT, summaries, action items
packages/sinks/       filesystem, stdout, webhook, Strata, TotalRecall
packages/kujo/        executable checks
.kujo/                specs and workflows
docs/                 setup, architecture, consent, recording, providers

Project status

Version 0.1.0 includes the complete file-first workflow, local capture, transcription providers, exporters, sinks, validation, Discord controls, and an experimental Discord-native receiver. The remaining production-readiness gate is live, multi-speaker Discord-native acceptance testing.

Documentation

• Getting started and CLI reference
• Recording on macOS
• Architecture
• Consent model
• Providers and Discord DAVE/E2EE
• Changelog

License

MIT © Robert DeVore — see LICENSE.