One-click recording
Mic and system audio as separate tracks, captured natively — both sides of any call app. No meeting bot.
BETAFree for noncommercial use · v0.1
Every meeting, remembered.
Nothing leaves your machine.
Sirina records your calls, transcribes them on your Mac's Neural Engine and summarizes them with a local AI model — all on your own computer. No bots in the meeting, no uploads, no account.
- 100% on-device
- Works offline
- Apple Silicon · macOS 14+
Recent
New recording
Statistics study group
3Transcribed on the Neural Engine — offline
Earlier this month
Thesis interview — participant 04
Export
4Markdown, text or copy
2Switch tabs
Overview · template: Research interview
Participant 04 describes how their household tracks energy use. They rely on monthly bills rather than live data, and say the main barrier to change is not knowing which appliances matter most.
3Summaries by your local AI model
Key points
- Checks the bill once a month; no smart meter app in use.
- Would act on per-appliance feedback if it were simple.
- Trusts neighbours' advice more than utility tips.
Follow-ups
Code quotes on "neighbour trust" theme.
00:12
Anna (You)
Thanks for joining. Could you walk me through how you keep track of your energy use at home?
00:24
Participant 04
Honestly, I just look at the bill when it comes. There's no app or anything — I wouldn't know where to start.
00:41
Anna (You)
What would make it easier to change something?
00:49
Participant 04
Knowing which appliances actually matter. My neighbour said the old freezer was the big one — and she was right.
3Timestamped transcript
What did they say would help them change habits?
Knowing which appliances matter most. They also trust advice from neighbours more than tips from the utility.
3Answers from your transcript
Ask about this recording…
Recording
0:00
1Audio saved straight to disk
2Stop — the transcript is nearly done
Nothing is streamed anywhere. Sirina transcribes on your Mac while you record, so only the last few minutes are left when you stop.
Local by design
Your conversations stay yours.
Recording, transcription, speaker separation and AI summaries all run on your computer. There is no Sirina server, no account and no telemetry — once the models are downloaded, pull the network cable and it keeps working.
- Audio files and the database live in one folder you control.
- Whisper runs on the Neural Engine via WhisperKit, with faster-whisper as a CPU fallback — no API keys.
- Summaries and Ask use your own model through Ollama.
- Prefer a cloud model? It's opt-in, and Sirina tells you clearly when data would leave the device.
Your machine
Record
Microphone + system audio
Save to disk
~/Library/Application Support
Transcribe
WhisperKit · Neural Engine
Summarize
Ollama · local LLM
0 bytessent to the cloud by default
Features
Everything a meeting leaves behind, organized.
On-device transcription
WhisperKit large-v3-turbo on the Neural Engine: 10 minutes of audio in about a minute. Detects the language and takes spelling hints for names.
Live captions & transcript
Optional live captions while you talk (macOS 26+). The transcript is finished during the call, so it's ready minutes after you stop.
Speakers & peopleBETA
Speaker splitting separates who spoke. Name a voice once and Sirina recognises it in later recordings.
Local AI summaries
Overview, key points and action items, written by the model you run in Ollama.
Ask your recordings
Chat with one meeting or all of them. Answers are grounded in your own transcripts.
Summary templates
Standard meeting, 1:1, standup — or write your own prompt structure and reuse it.
Tags & people
Color-coded tags and a people directory keep every recording easy to find.
Open formats
Export summaries and transcripts as Markdown or plain text. Your audio stays standard files.
Who it's for
For conversations you'd rather not upload.
Built for personal use, research, education and non-profits — the uses the PolyForm Noncommercial license welcomes.
Researchers & students
Interview participants without sending their voices to a third party. Keep ethics boards and consent forms happy.
Non-profits & communities
Minutes for board meetings, volunteer calls and member assemblies — without a per-seat SaaS bill.
Privacy-minded individuals
Remember what the doctor, the lawyer or your family actually said. Personal calls stay on your own disk.
Tinkerers & self-hosters
Pick your Ollama model, tune windows and threads, switch to faster-whisper — and read every line of the code.
PUBLIC BETA
Early, honest, and getting better every release.
Core recording and transcription are solid for daily use. Some parts are still moving — here's where things stand.
- Recording & transcriptionstable
- Summaries, Ask, templatesstable
- System audio capturenative
- Speaker splittingbeta
- Live captionsmacOS 26+
- Signed download (.dmg)build from source
- Windows & Linux buildsplanned
Source available. Read every line.
Sirina's source is public under the PolyForm Noncommercial License 1.0.0— free for personal, research, educational and non-profit use. Audit it, build it, improve it.
$ git clone https://github.com/st3v3y/sirina $ cd sirina $ ollama pull llama3.1:8b-instruct $ ./scripts/build-macos-app.sh ✓ speech-engine smoke test passed ✓ Sirina.app built · runs 100% local