Transcribe Discord calls on Mac without bots

Discord has evolved into the primary operating forum for open-source software maintainers, artificial intelligence research groups, decentralized protocol teams, and independent game studios. On any given evening, dozens of developers congregate in Discord voice channels and Stage rooms to debate Request for Comments (RFC) proposals, organize sprint goals, and review pull requests.

Despite the intensity of intellectual work happening inside these channels, documenting Discord conversations is notoriously painful. Traditional Discord recording bots require elevated server permissions, complex token provisioning, and awkward sound notifications. Even when configured, they merely spit out raw, multi-track audio files that require manual cloud transcription.

Oyma introduces a clean, client-side paradigm for macOS developers. By capturing audio directly from Apple’s ScreenCaptureKit pipeline, Oyma transcribes Discord voice channels on-device using local Whisper models, outputting structured Markdown records into your local vault without requiring server admin roles, bot tokens, or cloud transcription subscriptions.

The administrator permission dilemma with Discord bots

If you have ever tried capturing a community discussion on Discord with existing utilities (such as Craig or specialized server bots), you have experienced the friction:

1. The administrator role barrier: Conventional bots require Manage Server or full Administrator permissions to invite. If you are participating in a large community server, a client DAO, or an open-source project where you do not hold admin credentials, you simply cannot invite a recording bot.
2. Community security policies: Many server moderators explicitly ban third-party bots to protect servers against compromised tokens, webhook exploits, or spam raids.
3. Disruptive bot behavior: Conventional bots join the voice channel as a visible avatar and blast an automated audio chime ("Now recording") through the voice channel, breaking the conversational flow of informal discussions.
4. Disjointed audio dumps: Standard Discord bots export raw multi-track .flac or .wav archives onto external web servers. You are left to download gigabytes of audio files, pass them through a separate transcription tool, and format notes manually.

Operating outside the Discord API: OS-level audio interception

Oyma bypasses the Discord Developer Portal, bot tokens, and server permission hierarchies entirely by operating at the macOS hardware interface layer:

┌────────────────────────────────────────────────────────────────────────┐
│                   DISCORD DESKTOP AUDIO TOPOLOGY                       │
├────────────────────────────────────────────────────────────────────────┤
│  Discord Voice Channel / Stage                                         │
│       │                                                                │
│       ▼                                                                │
│  Discord Electron Client Audio Output                                  │
│       │                                                                │
│       ▼                                                                │
│  macOS CoreAudio Mixer ───► Local Speaker / Headphone Stream           │
│       │                                                                │
│       ├─────────────────────────┐                                      │
│       ▼                         ▼                                      │
│  [ScreenCaptureKit Output]  [Local Mic Input]                          │
│       │                         │                                      │
│       └───────────┬─────────────┘                                      │
│                   ▼                                                    │
│      [Apple Silicon Whisper Model] (On-Device Neural Engine)           │
│                   │                                                    │
│                   ▼                                                    │
│      Sovereign Markdown Output: vault/meetings/discord-*.md            │
└────────────────────────────────────────────────────────────────────────┘

Because capture occurs directly on your workstation:

• You can capture voice discussions on any server, regardless of your role or permissions.
• Direct message (DM) voice calls and private group voice chats are supported with zero special configuration.
• No third-party OAuth bot token is stored or exposed.

Reactive CoreAudio channel detection

When you connect to a Discord voice channel or join an open Stage session, the Discord macOS client opens a CoreAudio output stream. Oyma monitors these audio state events in src/main/meeting/meeting-detector.ts and alerts you with a subtle overlay:

Discord voice channel active.
[Record Voice Channel]   [Dismiss]

Capture requires your manual confirmation. You retain absolute control over whether an informal gaming session or a critical engineering design review gets recorded.

Handling fluid community voice dynamics

Unlike structured Zoom meetings where participants join simultaneously and stay seated, Discord voice channels are fluid. Contributors drop in, ask a question, share an opinion, and disconnect.

Oyma handles these dynamic sessions effortlessly:

• Continuous local stream ingestion: The audio engine maintains seamless recording continuity even as voice participants connect and disconnect across hours of discussion.
• Multilingual Whisper processing: International open-source communities often switch between English, French, Japanese, or German. Local Whisper models detect language shifts automatically across 13 supported languages.
• Zero cloud token costs: Transcribe long four-hour developer hackathons or community office hours without worrying about runaway per-minute cloud API fees.

Open-source community call Markdown format

When the voice session wraps up, Oyma synthesizes the dialogue into an organized Markdown document ready for commit into your project’s repository:

---
title: "RFC Discussion: Distributed Cache Invalidation"
date: 2026-10-08
platform: Discord
guild: "Rust Tooling Collective"
channel: "voice-architecture-a"
tags: [discord, oss, rfc, rust, caching]
---

## Summary of Discussion
Reviewed proposed RFC #412 for asynchronous cache invalidation over pub/sub. Addressed race conditions during node partitions and reached consensus on event sequencing.

## Technical Consensus Points
- Sequencing mechanism: Adopt Lamport timestamps embedded in cache invalidation payloads rather than relying on synchronized system clocks.
- Partition handling: If a worker node detects a disconnect from the pub/sub broker for >2000ms, it must invalidate all local L1 cache entries automatically.
- Wire format: Reject JSON in favor of concise Protocol Buffers to reduce serialization overhead during high-frequency write storms.

## Community Action Items
- [ ] @alx_dev: Update RFC #412 text with the agreed Lamport timestamp specification
- [ ] @sophia_m: Benchmark Protobuf serialization latency against Cap'n Proto on staging
- [ ] Community: Next voice sync scheduled for Tuesday in #voice-architecture-a

Store this file directly in your Obsidian vault or git-tracked project repo, linking to other design docs using standard [[wikilinks]].

Sovereign privacy for pseudonymous contributors

In open-source, crypto, and research communities, contributors frequently operate under pseudonyms and value data privacy above all else. Routing live community audio through consumer cloud notetakers risks exposing voice prints, biometric telemetry, and private metadata to surveillance pipelines.

With Oyma, all speech processing occurs locally on your Mac’s Apple Silicon hardware. You can verify through the built-in Egress Log that zero audio bytes or transcripts are transmitted across the network.

Take control of your community discussions. Capture Discord voice channels with total local data sovereignty and plain Markdown files today.