Cicada
Cicada is an Apache-2.0 meeting transcription and translation tool for Mac, Windows, Linux and Android. It captures the system audio of a call or livestream, shows live bilingual subtitles, writes structured notes with summaries and to-dos, and on Apple silicon Macs runs the recognition on the device. On other platforms it needs an AI service key, so audio leaves the machine.
What does Cicada do?
Cicada's README says: "Every call. Every stream. No language barrier. Understand, speak back, and remember everything." It captures the system audio of a meeting or livestream, writes a live bilingual transcript and subtitles with automatic language detection, and keeps structured notes beside the call: summaries, topics, conclusions, to-dos and fishbone diagrams.
It also offers simultaneous interpreting. You hold a talk key and speak, and your words are translated and delivered into the call with an optional voice clone. On Mac it identifies speakers and keeps a voice memory across meetings.
Key facts
- Licence: Apache-2.0 on the code. The bundled BlackHole audio driver is GPL, and the bundled voiceprint model is CC-BY-4.0, per the README.
- On-device recognition on Apple silicon Macs through Apple's
SpeechAnalyzerandSpeechTranscriber; the README says "all speech recognition runs on device, so audio never leaves your computer." - On Intel Macs, Windows, Linux and Android, "speech recognition runs in the cloud and requires an AI service key; audio is sent in segments to Ling's Ming or your chosen service."
- Translation: Ling
Ling-3.1-flashor Google Translate as a fallback. - Interpreting: Ling
Ming-3.0-audio-flashor on-device recognition chains. - Speaker identification uses the bundled WeSpeaker ResNet34-LM model through Core ML on Mac.
- Exports: Markdown, Word and SRT.
- Share notes with an expiration date and a password.
- Platforms: Mac 14.4 and later (Apple silicon and Intel), Windows 10 2004 and later (x64 and ARM64), Ubuntu 22.04 and 24.04 desktop, Android 10 and later.
What does it replace, and where does it fall short?
Cicada replaces Otter.ai and Fireflies.ai for a user who wants structured meeting notes from calls and livestreams, with the option to run the recognition on the device on an Apple silicon Mac. It is listed on the open-source alternatives to Otter.ai page beside Meetily, Buzz, Vibe and Scribe.
Where it falls short: only Apple silicon Macs get the fully local path. The README states that Intel Macs, Windows, Linux and Android need an AI service key, so the audio goes to a cloud service. The notes window is visible in a screen share on Linux. Call and meeting apps on Android block internal audio capture, and the README's workaround is to use speaker mode. Recording the user's own speech on their side of the call is on the roadmap, not shipped. The builds are not code-signed or notarised.
How does Cicada run?
As a desktop app on Mac, Windows and Linux, and as an Android app on phones running Android 10 and later. On an Apple silicon Mac the recognition runs with Apple's SpeechAnalyzer and the bundled WeSpeaker model for speaker identification, both on the device. On every other platform the audio is sent in segments to Ling's Ming model or another AI service of your choice, and you set the key in the settings.
The bundled BlackHole driver captures system audio on Mac so the app can hear both sides of a call. Builds are available from GitHub releases and the project's website.
Who is Cicada for?
A reader who wants a live bilingual transcript, automatic translation and structured meeting notes on calls and livestreams, and who is on an Apple silicon Mac for the local-only path. A reader on Windows, Linux or Android who wants a fully offline tool should look at Meetily or Scribe instead, because Cicada's non-Mac path needs a cloud key.
What limits does the README state?
- Intel Macs, Windows, Linux and Android "require cloud-based speech recognition via API key."
- "Call/meeting apps on Android block internal audio capture", and the workaround listed is speaker mode.
- "Notes window visible in screen sharing on Linux."
- "Recording user's own speech not yet supported."
- Builds are not code-signed or notarised; the roadmap lists signing and notarisation as a priority.
Questions people ask
Is Cicada open source?
Yes. The repository is Apache-2.0 on the code, an OSI-approved licence with a patent grant. The README notes that the bundled BlackHole audio driver is GPL and the bundled voiceprint model is CC-BY-4.0, so a downstream user who ships those bundled files inherits those terms for those files.
Does Cicada send my audio to the cloud?
On an Apple silicon Mac, no. The README says the recognition runs on the device through Apple's own SpeechAnalyzer and the bundled WeSpeaker model. On an Intel Mac, Windows, Linux or Android, yes: the README states that speech recognition on those platforms runs in the cloud and needs a service key, and the audio is sent in segments to Ling's Ming or the service you chose.
Can Cicada translate the call while it is happening?
Yes. The README lists live bilingual transcripts with automatic language detection, and a simultaneous interpreting mode where you hold a talk key, speak, and the translated voice is delivered into the meeting, with an optional voice clone. Translation goes through Ling Ling-3.1-flash or Google Translate as a fallback.
How does Cicada compare with Otter.ai?
Otter.ai is a hosted service that stores recordings and notes on its servers on a subscription. Cicada runs on your own devices, keeps notes on your machine, and on an Apple silicon Mac runs the recognition on the device too. Its structured output (summaries, topics, conclusions, to-dos and fishbone diagrams) and bilingual transcripts are the Otter-style notes, and its interpreting mode adds something Otter does not offer. The open-source alternatives to Otter.ai page sets it beside Meetily, Buzz, Vibe and Scribe.
Does Cicada work on Zoom, Teams and Google Meet?
Yes, through system-audio capture. The README lists speaker identification and voice memory on Mac for Zoom, Teams and Meet, and the general audio capture works on calls that play through the system. On Android, calling apps block internal audio capture by platform policy, so the README's workaround is to put the call on speaker.
Sources
- Cicada README and LICENSE: github.com/wquguru/cicada, read 2026-11-16. Code licensed Apache-2.0, bundled BlackHole driver GPL, bundled voiceprint model CC-BY-4.0.
More meeting transcription
A meeting assistant under MIT that records your microphone and system audio together, transcribes live on your machine with Whisper or Parakeet, and summarises with a language model you choose.
A desktop app under MIT that transcribes and translates audio and video files, YouTube links and a live microphone offline with Whisper, with speaker identification and SRT, VTT and TXT export.
An offline transcription app under MIT that records system audio and the microphone, supports Whisper, Parakeet and Nemotron models, labels speakers, and exports SRT, VTT, TXT, HTML, PDF, JSON and DOCX.
An MIT self-hosted transcription server that runs Whisper large-v3 and pyannote speaker diarization on an NVIDIA GPU, with a web UI, REST API and CLI, tested on an RTX 3060 Ti.
An MIT local-first meeting notes app for Apple Silicon Macs that records a Google Meet, Zoom or Microsoft Teams call through a browser extension, transcribes with Whisper Large V3 Turbo and summarises with Ollama, with no bot joining the call.
Added November 16, 2026. Every claim above comes from the project's README, LICENSE or model card, read on November 16, 2026, or from the GitHub API on the date shown in the panel. Found an error? Write to reveneau@licheo.com and it is fixed in the next weekly pass. Repository: github.com/wquguru/cicada.
Want Cicada running on your own servers?
Reveneau sets it up on your hardware, connects it to your product, and builds the software around it.