Skip to content
Self-hosted · No audio leaves your machine

Your Discord
meetings, transcribed
and summarized.

Parley joins your voice channel, transcribes every speaker on your own machine, and drops structured AI meeting notes — decisions, action items, talk-time — straight into a Discord thread.

Quickstart → View source
ISC · Free faster-whisper Gemini · Ollama · OpenAI SQLite FTS5
LIVE // VOICE CAPTURE Rec
PP
PixelPaladin

Build is green on main. Frame pacing fix is in.

whisper · local · per-speaker → thread

✕ Cloud meeting tools

  • ✕ Per-seat monthly pricing
  • ✕ Your audio uploaded to their servers
  • ✕ Vendor decides retention & access
  • ✕ Yet another SaaS account

◆ Parley

  • ✓ Free and open source (ISC)
  • ✓ Audio transcribed on your own machine
  • ✓ You own the data and the database
  • ✓ Runs on a Raspberry Pi or a GPU box
// 01 — Demo

See it in action

Scroll to run a live meeting — Parley transcribes each speaker, then writes the notes.

Pixelforge // #launch-week Rec
Live transcript
PP
PixelPaladin

Build is green on main. Frame pacing fix is in.

RR
RespawnRita

Save corruption bug is fixed — but co-op desyncs on level 3.

LG
LootGoblin

What if we add one more boss before launch?

PP
PixelPaladin

No. We freeze scope today or we miss Friday.

NN
NoScopeNova

Trailer drops Wednesday, wishlists are climbing.

RR
RespawnRita

Then co-op has to be cut or delayed — it is not stable.

PP
PixelPaladin

Agreed. Ship solo Friday, co-op as a week-1 patch.

Meeting ended — generating notes
P
Parley BOT posted meeting notes

TL;DR

Pixelforge ships solo mode this Friday. Co-op is cut from launch and moves to a week-1 patch due to a level-3 desync. Trailer goes live Wednesday.

Decisions

  • ▸Launch date locked: Friday.
  • ▸Co-op mode cut from v1 — ships as a week-1 patch.
  • ▸Scope is frozen as of today — no new content.

Action items

PP PixelPaladin
  • ☐Tag the release branch and freeze scope
  • ☐Prep the week-1 co-op patch plan
RR RespawnRita
  • ☐File the level-3 desync repro
  • ☐Sign off on the solo-mode build
LG LootGoblin
  • ☐Shelve the extra-boss design doc for v1.1
NN NoScopeNova
  • ☐Schedule the trailer for Wednesday
  • ☐Queue the launch-day announcement

Talk-time

PixelPaladin 38%
RespawnRita 27%
NoScopeNova 20%
LootGoblin 15%
// 02 — Features

Everything a meeting bot should be

Built for self-hosting, privacy, and the way Discord actually works.

01

Per-speaker attribution

Discord gives each user a separate audio stream, so every line is attributed to the right person — not guessed by an ML model.

02

Structured AI notes

TL;DR, decisions, open questions, and action items grouped by the person responsible — plus per-speaker talk-time.

03

Pluggable summarizer

Google Gemini (free tier), any OpenAI-compatible endpoint, or fully-offline Ollama — switch per server, no restart.

04

Local speech-to-text

A warm faster-whisper sidecar runs on your hardware. Pick a model from tiny to large-v3-turbo.

05

Runs anywhere

Node's built-in node:sqlite means no native build. Works on a Raspberry Pi or a GPU server alike.

06

Searchable history

Full-text /search across every past meeting, backed by SQLite FTS5. Plus /history and /summary.

07

Auto join / leave

Joins when 2+ people are talking, leaves when the room empties. Shows [REC] in its nickname while recording.

08

Concurrent meetings

Records multiple channels and servers at once — keyed by guild and channel, with no global single-recording limit.

// 03 — Pipeline

How it works

Two processes on your hardware. Audio never leaves the box.

  1. 01

    Capture

    Per-user PCM audio is captured from the voice channel — one isolated track per speaker.

  2. 02

    Transcribe

    Each track is fed to the warm faster-whisper sidecar running locally on your machine.

  3. 03

    Summarize

    The merged, speaker-labeled transcript goes to your chosen LLM — Gemini, Ollama, or OpenAI-compatible.

  4. 04

    Deliver

    Structured notes are posted to a Discord thread. Audio is deleted after transcription.

// 04 — Privacy

Private by architecture

  • ✓Audio is transcribed on the machine running the bot — nothing is uploaded to Parley.
  • ✓Only the final transcript text is sent to the summarizer you choose. With Ollama, nothing leaves your network at all.
  • ✓The bot shows [REC] in its nickname whenever a recording is active, so every member can see it.

Recording voices is subject to consent laws that vary by jurisdiction. You are responsible for obtaining consent from all participants.

// 05 — Quickstart

Up and running in minutes

Two processes: the Node bot and the Python transcription sidecar.

~/parley 
git clone https://github.com/SakethKanchi/parley.git
cd parley
npm install

API keys live in .env only — never typed into Discord. Full docs in the README.