# Synopsule

> A privacy-first transcription app for Mac and iPhone that leaves you with notes you will actually re-read: a clean transcript with who said what, the full audio kept on-device so you can replay any moment, and notes saved as open Markdown any AI agent can read. It records your conversation, transcribes it locally with OpenAI Whisper running on-device (a model is bundled, so it works on first launch), labels every speaker after the recording finalizes, recognizes returning voices across recordings on-device, and only summarizes when you press the button — so nothing leaves your device unless you ask. One app covers both platforms. Optimized for Apple silicon. macOS 15 Sequoia or iOS 17 or later.

Live on the App Store as a $1.99 one-time universal-app purchase covering both Mac and iPhone (https://apps.apple.com/us/app/synopsule/id6773112941). **Current version: 1.2.** Headline additions since 1.0: a built-in MCP (Model Context Protocol) server on Mac so Claude Desktop, Codex, Claude Code, and ChatGPT can read, search, and summarize your transcripts directly from inside the app (see the "MCP integrations" section below); structured, speaker-aware **templated summaries** (one-on-one, standup, sales call, interview, lecture, clinical SOAP, legal, user research, and a general default) you pick per recording; a **Notes mode** for time-anchored notes that tap-to-seek and fuse into the summary as authoritative context; **surfaced diarization** with "Recognized" chips when a saved voice is matched; **open-data export** straight to Obsidian and Apple Notes; **on-device summaries via Apple Intelligence** (Apple's Foundation Models — no key, no network, on Macs and iPhones running macOS 26 / iOS 26 on Apple-Intelligence-eligible hardware); **live corrections** — speaker IDs and text fixes you make while recording survive the finalize and steer the summary; and **labelable flags** you can drop and name mid-recording, surfaced in a tabbed reader companion pane (Summary / Notes / Speakers / Flags). AI summaries are optional and offered four ways: **bring-your-own API key** (free, transcript text goes direct to OpenAI / Anthropic / Google Gemini), a **local model via Ollama on Mac** (fully on-device — nothing leaves at all, no key or subscription needed), **on-device Apple Intelligence** (Mac and iPhone, on eligible hardware — also nothing leaves), or **Synopsule Pro** ($4.99/month or $39.99/year, covering both devices) for one-tap summaries with no key setup.

## Site

- [Home](https://synopsule.com/): product overview, screenshots, pricing, FAQ
- [Privacy policy](https://synopsule.com/privacy/): the full data-handling policy
- [App Store — universal, Mac + iPhone](https://apps.apple.com/us/app/synopsule/id6773112941): $1.99 one-time purchase
- Support: rxv2dev@gmail.com

## Platforms (one app, two devices)

- **Mac** — sees the most of any conversation. Captures your microphone and your Mac's system audio in separate lanes, so it records both sides of a Zoom / Meet / Teams call with no bot in the room. Export includes PDF. Roomy workspace: library, transcript, and a companion pane that tabs through Summary, Notes, Speakers, and Flags. macOS 15 Sequoia+.
- **iPhone** — the same studio in your pocket. Records the room through the microphone, transcribing live. Built for in-person meetings, interviews, lectures, table talk, and voice notes. iOS 17+, iPhone only. Microphone only (no system-audio capture); export covers everything except PDF.
- **Both** — on-device Whisper transcription (works on first launch from a bundled model), automatic speaker labels you rename once, on-device recognition of returning voices across recordings, tap-to-play playback synced to the transcript, full-text search across every transcript, and opt-in AI summaries. A purchase or Pro subscription on either device covers the other.

## What it does

- **Records the conversation.** On Mac, captures your mic and the Mac's system audio in separate lanes (Apple's public Core Audio process-tap APIs, macOS 14.4+) — no screen recording, no virtual audio driver, no bot. On iPhone, records the room through the mic. Works for Zoom, Google Meet, Microsoft Teams (on Mac), and in-person meetings, podcasts, interviews, and lectures (both).
- **Transcribes locally with Whisper.** OpenAI Whisper running on-device. A model is bundled in the app, so transcription works on the very first launch with no download and no connection. Download a larger model size from tiny (~39 MB) to large (~1.5 GB) when you want more accuracy; a balanced ~626 MB model is a good default. Keep several downloaded models and switch between them instantly with no redownload. Transcripts appear live as you talk.
- **Labels speakers automatically, and recognizes returning voices.** After the recording finalizes, on-device diarization identifies distinct voices and tags every line. Rename "Speaker 2" to "Maria" once and it propagates everywhere — transcript, search, exports, and a speaker directory that tracks people across recordings. Synopsule also builds an on-device voiceprint per named person (using a bundled CAM++ neural voice model) and **auto-labels that person in future recordings**; correcting a wrong guess trains the matcher. When a saved voice is matched, the line carries a **"Recognized" chip** so you can see at a glance who was auto-identified versus freshly labeled. Match sensitivity is tunable, and voices can be merged or forgotten — a persona-consistent merge keeps one person's lines together. Voiceprints never leave the device. Works on both Mac and iPhone. Tell it the speaker count beforehand for sharper diarization.
- **Optional AI summaries, with templates.** Press Summarize and pick a template — general, one-on-one, standup, sales call, interview, lecture, clinical SOAP, legal, or user research — and get a structured, speaker-aware recap with action items and follow-ups shaped to that kind of conversation. All templates are free; the specialized ones (Clinical SOAP, Legal) read best with a cloud model. Choose where it runs: BYO OpenAI / Anthropic / Google Gemini key (transcript text goes direct, you pay only the provider); on a Mac, a local model via Ollama (summary generated on-device — nothing leaves at all, no key or subscription); **on-device Apple Intelligence** on Macs and iPhones with eligible hardware (macOS 26 / iOS 26 — also fully private, no key, no network); or Synopsule Pro (transcript text routes through Synopsule's worker to OpenAI). The *audio* never leaves your device. AI is opt-in — if you never press Summarize, nothing leaves at all, and a consent sheet names the destination before any text is sent.
- **Notes mode.** Jot time-anchored notes while you record or read; tap a note to seek straight to that moment in the audio, and your notes fuse into the AI summary as authoritative context (the model treats what you wrote as ground truth).
- **Live corrections that survive finalization.** Fix a misheard word or reassign a speaker while you're still recording, and those edits carry through the wholesale finalize pass instead of being overwritten — and they inform the summary. Speaker IDs you set live always re-apply; text fixes re-apply conservatively.
- **Flag the moments that matter.** Drop a flag mid-recording and give it a label; flags become tappable jump points and live in their own tab/sheet in the reader, alongside the transcript.
- **Plays back, synced.** Tap any line to jump to that moment; the active line follows as the audio plays. The audio stays on the device — keep it or delete it.
- **Exports anywhere, including your notes app.** Word (.docx), Markdown, plain text, SRT, VTT subtitles, interactive HTML on both platforms, plus PDF on Mac — and one-tap **open-data export straight into Obsidian or Apple Notes** so a transcript lands where you already think. Toggle timestamps, speaker labels, grouping by speaker, and whether the AI summary travels with the file. It's plain, portable text — no lock-in.
- **Searches everything.** Full-text search matches across every transcript, with `speaker:` filters to jump to a voice.
- **Built for long sessions.** Live transcription is incremental; playback streams from disk. Hour-plus recordings stay smooth.

## MCP integrations (Mac)

Synopsule includes a built-in **MCP (Model Context Protocol)** server (shipped in 1.1) so AI assistants can read, search, and pull from your local transcript library on demand — no copy-paste, no upload, no separate helper to install. The server is the Synopsule app itself, launched with the `--mcp` flag, so it always runs at the same version as the app and lives inside the app's sandbox (local-only, read-only).

- **Supported clients (Mac):** Claude Desktop (one-click install of `Synopsule Transcripts.mcpb`), Codex app / CLI / IDE (one config covers all three via `~/.codex/config.toml`), Claude Code (`claude mcp add synopsule -- "/Applications/Synopsule.app/Contents/MacOS/Synopsule" --mcp`), and ChatGPT (drag a Markdown export into the chat, or expose the server through a temporary HTTPS tunnel like cloudflared since ChatGPT custom connectors require public HTTPS).
- **Setup UI:** Synopsule → Settings → Connect to AI Apps. Each supported client has a one-click install (Claude Desktop, Codex) or copy-config button with the absolute path pre-filled. Manual command: `/Applications/Synopsule.app/Contents/MacOS/Synopsule --mcp`.
- **Tools exposed to the AI:** `list_transcripts` (recent recordings with id, date, duration, speakers), `search_transcripts` (full-text with timestamped speaker excerpts), `get_transcript` (full Markdown for one recording, auto-chunked when long), `get_summary` (just the AI recap — summary, action items, follow-ups).
- **What it serves:** a Markdown mirror of each finalized recording (YAML frontmatter + a `## Transcript` section) plus an `index.json` catalog. Drafts, voice embeddings, speaker personas, and audio files are never mirrored or exposed. An optional human-browsable export folder (Settings → Connect to AI Apps → Export folder) keeps an identical copy in any Finder / iCloud Drive / Obsidian location you choose.
- **iPhone:** The iOS app surfaces the same Markdown mirror in Files → On My iPhone → Synopsule (Settings → Storage → Show in Files app). Share into ChatGPT or Claude iOS apps from there, or put the Mac's export folder in iCloud Drive to read Mac recordings on the go.
- **Useful queries to suggest the user try:** "Search my transcripts for the pricing discussion and summarize what we decided", "What action items came out of yesterday's standup?", "Find every mention of <person> in the last month and list them with timestamps."

## Privacy architecture (the part people miss)

- **Raw audio** is captured to a temporary file on the device, processed by local Whisper, then **deleted after final processing**. A playback copy is kept locally so you can replay alongside the transcript — you can delete it any time.
- **Transcripts and metadata** live in the app's sandboxed storage. Nothing syncs.
- **API keys** (if you bring your own) are stored in the device Keychain — Synopsule itself never sees them.
- **Speaker voiceprints** (the acoustic fingerprints used to recognize returning voices) are computed and stored on-device only; they are never uploaded. The on-device Ollama summary path (Mac) keeps even the transcript text on the machine.
- **Analytics** are anonymous and opt-out-able. Audio and transcript content are never sent to analytics.
- **The cloud-summary path** (Synopsule Pro) sends only the transcript text to a Cloudflare Worker which validates the App Store subscription and forwards to OpenAI. The Worker doesn't persist transcript text after the request completes.

## Who it's for

- Operators and PMs running long internal syncs who want notes without inviting a bot into the call (Mac)
- Anyone capturing in-person conversations, interviews, lectures, or voice notes on the go (iPhone)
- Sales / customer success folks who need accurate call transcripts but can't ship customer audio to third-party SaaS
- Founders / consultants under NDA where the recording can't legally leave the device
- Journalists and researchers conducting interviews where consent only covers on-device processing

## What it isn't

- Not a Zoom bot (no one joins the meeting; you record on your own device)
- Not a SaaS — no account, no cloud library, no team workspace
- Not a real-time translator (transcription only, single-language)
- A one-time $1.99 universal-app purchase on the App Store covering both Mac and iPhone (not free; no separate ad-supported tier)
- On iPhone, not a remote-call recorder — it records the mic, so use the Mac for both sides of a call

## System requirements

- macOS 15.0 Sequoia or later, or iOS 17.0 or later (iPhone only)
- Optimized for Apple silicon (M-series Macs; A-series iPhones)
- A Whisper model is bundled in the app, so it transcribes on first launch; optionally download a larger 39 MB–1.5 GB model for more accuracy (your choice of size, kept for instant switching)
- Optional on-device AI summaries on Mac require Ollama (a free local model server) installed separately

## License & terms

- App Store standard EULA: https://www.apple.com/legal/internet-services/itunes/dev/stdeula/
- Privacy policy: https://synopsule.com/privacy/

## Contact

- General + support: rxv2dev@gmail.com
- Built by Sunny Nagra (indie)