Synogram records, transcribes, and summarizes meetings using your own OpenAI key — locally on your device. No subscription. Nobody in the middle holding your audio.
macOS 14.4+ · iOS 17.0+ · Bring your OpenAI or OpenRouter key
The team aligned on three priorities for Q3: shipping the BYOK flow, cutting transcription latency for sessions over 30 minutes, and opening Mac TestFlight to a wider cohort.
Mai 00:12 — Should we ship the BYOK toggle this sprint or wait until cloud auth is GA?
Ken 00:18 — Ship it this sprint. Cloud is a strict superset; we don't gain anything by waiting.
Ranatchai 00:24 — Agreed. Privacy story is sharper if BYOK lands first.
Above: Synogram on macOS · simulated meeting
Otter, Fathom, Granola — month after month after month. And while the bill is annoying, the deeper cost is harder to see: every call you've ever recorded is now sitting on someone else's server, indexed against terms you didn't read.
You don't need another SaaS. You need the audio to stay where it was recorded — on your Mac, your iPhone — and the AI to do its work without anyone in the middle.
Your Mac records. Your key transcribes. Your key summarizes. Nothing sits in the middle.
Mac captures both your mic and system audio — Zoom, Meet, Teams, anything playing. iPhone records the room. Audio is written straight to a local file. Nothing leaves your device yet.
recording.caf · written locally
The audio chunks are sent directly from your laptop to OpenAI using your API key. Speakers are diarized, timestamps attached, language auto-detected — Thai, Japanese, English, more.
A · 00:12 Should we ship BYOK this sprint?
B · 00:18 Yes — cloud is a strict superset.
Pick a template — Standup, UX Research, Sales, Discovery, Retro, or General — and the transcript becomes a structured summary with action items, decisions, and per-speaker notes.
— Profile transcription on 60-min recording (Ken)
— Open TestFlight to next 50 sign-ups (Ranatchai)
In BYOK mode, your laptop talks directly to OpenAI. Synogram has no proxy, no server, no logs, no copy of your audio. We literally cannot read your meetings — we don't run anything that touches them.
Most meeting tools were designed cloud-first — you upload your audio, they do the work, they keep a copy. Synogram is designed the other way around: local-first, with the cloud as an optional convenience you control.
Every meeting — audio, transcript, summary, your notes, your task progress — lives in a local SQLite database (GRDB) inside the app's sandbox. Back it up to Time Machine. Move it to a new Mac. Delete a recording and it's gone for real.
Your OpenAI and OpenRouter keys are stored in the platform Keychain — Secure Enclave-backed on iPhone, the same place Safari keeps your saved passwords on Mac. They never appear in plaintext on disk and never reach Synogram's servers (we don't have any in BYOK mode).
When transcription runs, the audio chunk goes straight to api.openai.com over TLS, authenticated with your key. Synogram doesn't proxy it, doesn't log it, doesn't see it. The network trace shows two endpoints: your machine and OpenAI's.
If you want billing convenience instead of managing API keys, you can sign in to Synogram Cloud — but you have to choose it. The app ships in BYOK mode. No background sync. No "anonymized telemetry." No surprise upload.
Each template is purpose-built — different headings, different angles, different signals to pull out. Outputs stay in the original language, no translation.
Executive summary, decisions, action items, per-attendee notes. The safe default when you're not sure which template fits.
Yesterday / Today / Blockers per person. Plus risks, decisions, and a clean action item list — what daily standup notes should always look like.
Participant snapshot, insights, pain points, verbatim quotes, validated hypotheses, surprises, and follow-up questions.
Account snapshot, pain points, use cases, objections, buying signals, decision process, pricing notes, next steps.
Who they are, job-to-be-done, current workflow, what works, friction, workarounds, tools in play, opportunities, open questions.
TL;DR, what went well, what didn't, lessons, action items, experiments to try next sprint. Per-speaker notes preserved.
Not the right shape? Open the summary, ask a follow-up question in plain language, and Synogram refines it grounded in the transcript — same key, same privacy posture.
You picked the key. Pick the model, the prompt, the speaker names. Synogram gets out of the way.
You're paying OpenAI directly. Use any model worth the tokens — and switch when a sharper one ships next month, without losing the version that's already in your notes.
Ask questions in plain language and Synogram answers from this meeting — grounded in the actual transcript and summary, with line-level citations you can click into.
Mai — Ship BYOK toggle this sprint. Cloud is a strict superset.
Ken — Profile transcription on a 60-min recording; target sub-30s per 10 min.
Ranatchai — Open Mac TestFlight to next 50 sign-ups.
"Make it shorter. Group action items by owner. Cut the preamble."
Tell Synogram what to change in plain language — "shorter," "group by owner," "drop the preamble," "in Thai" — and it rewrites the summary. Every version is saved. Revert without losing your work.
OpenAI's diarization model splits voices automatically. Tag a speaker once — Mai, Ken, Ranatchai — and Synogram retroactively colors every line they said, in any language the model heard.
ลุยเลยมั้ย — เราจะ ship BYOK toggle ทันสปรินต์นี้ไหม?
ลุยเลยครับ — Cloud เป็น superset อยู่แล้ว ไม่ได้เพิ่ม risk อะไร
Applies to all 47 lines from Speaker B in this meeting.
เห็นด้วย — privacy story จะ sharp ขึ้นถ้า BYOK land ก่อน
ScreenCaptureKit pulls your microphone and the system audio coming from Zoom, Meet, Teams, FaceTime, anything playing — without screen recording, without a virtual driver, without a kext.
Background audio mode means the recording continues whether the phone is locked, in your pocket, or you've jumped into another app — and a Live Activity on the Lock Screen lets you stop with one tap.
Honest numbers. You pay OpenAI directly for the transcription and summarization tokens you actually use. Synogram charges nothing in BYOK mode — there's no per-seat fee, no recurring bill, nothing to cancel.
Pay OpenAI for usage. Nothing to Synogram.
Estimate: gpt-4o-transcribe at $0.006/min · ~$0.36/hr audio · plus a few cents per summary. Your actual cost depends on length and model choice.
Monthly fee. Their AI. Their cloud.
Pricing as of mid-2025 from each vendor's site. Annual plans roughly 20% lower; per-seat for teams.
Synogram Cloud is a single magic-link sign-in. We route your transcription and summarization through Vercel AI Gateway and bill you for what you use — no monthly subscription, no per-seat fee. Same privacy stance: zero data retention on the gateway, and you can switch back to BYOK any time.
Off the SaaS treadmill. Off someone else's server. Onto your Mac, where they always should have been.
No signup · No subscription · Your key, your data