Oats - Free open-source and on device meeting notetaker
by•
Oats is an AI meeting note-taking tool that is completely open, local, and free. Designed to not get in the way of your meetings, no bots, no subscription needed when running locally with on-device LLM. Available on both macOS and Windows.
Enhanced transcription, multi-language support, speaker recognition, assessment, coaching, auto-tracking follow ups and more features via ariso.ai cloud backend.
Replies
Does Oats keep the original audio along with the transcript?
@avery_thompson2 yep, you can playback the original audio. It also supports voice profile to for speaker matching
What happens if I close my laptop right after a meeting?
@carter_garcia I think that would pause/stop the recording/notes
@carter_garcia The audio files are queued, so you'll see an indicator saying it's ready to be processed when you re-open it.
This solves one of my biggest issues with meeting tools. how accurate is the local transcription compared to cloud based tools?
@colton_drake the local models do a decent job but honestly the cloud backend models is a little better but mostly for longer meetings, that's why we also provide a cloud backend ourselves, you can create a free account on ariso.ai to use the cloud backend.
@laura_cruickshanks love to hear your feedback especially on what features you hate
@laura_cruickshanks Thanks!!!
Open source + local AI is a great combo. can users swap in their own local models?
@james_carter35 Yes we plan to add support for user's own models soon
fyi @unplug_eth0
@james_carter35 Once this issue is fixed, you will be able to swap local model or use your favourite cloud models for local backend.
No meeting bot is a really nice touch. does Oats work smoothly with Zoom, Meet and Teams?
@barnaby_lloyd Oats records from computer microphone and speakers, Zoom, Meet, Teams, meeting room, noisy coffee shop, airport gate, Oats truly benefits from high quality audio devices and handing echo well.
Respect for the privacy and no-subscription approach!
@jackdonovan thank you!
The no-bot-in-the-call part is what stands out to me. I build voice AI for daily check-in calls with older adults, and the thing that consistently bites us is not the LLM, it is the front of the pipeline: soft or slurred speech, hearing aids, a TV going in the background. Which on-device ASR did you settle on for the local path, and how much accuracy do you give up versus the Ariso cloud route on noisy audio? Also curious whether speaker recognition is cloud-only by design, or if local diarization is on the roadmap.
@igorgurovich noise canceling does a pretty decent job most of the time, and even though the audio/transcript may not be perfect, the AI models can do a great job at summarizing the notes!
@igorgurovich Ariso cloud version will always be better. We went for an optimization on both cost and effectiveness for the local models. If there's some other models you'd like it to use you can always open a pull request on our github repo for Oats to add them!
Congrats @maxheckel how much CPU/RAM does it typically use while recording? curious how lightweight it is on older laptops.
@imogen_wallace Right now it requires around 2GB of ram to be effective. CPU shouldn't matter as much and should be fine with any modern laptop. If you use the cloud backend (which is also free), you can have a total potato of a laptop and still have it work effectively.
left a review above - the no-bot, fully local angle is what stands out here. curious how the diarization/speaker recognition holds up locally when the model is small enough to actually run on a laptop, that's usually where local AI notetakers fall behind cloud ones.
@galdayan diarization/speaker recognization for local backend is indeed a challenge b/c of the performance of local model. I’ve tried several STT models and Oats use the best I’ve tried.
remote model is indeed better and Ariso backend provides more features including speaker profile and auto matching. give it a try and share us your feedback!
@unplug_eth0 makes sense, that's a fair tradeoff to be upfront about rather than overselling the local mode. good to know Ariso is there for whoever needs the speaker-profile matching. will give the local backend a real try on some messier multi-speaker recordings and report back what I find.
@galdayan that would be great, just comment or provide feedback in https://github.com/ariso-ai/oats The good news is I've talked to lots of top AI labs who pioneers in speech model, speaker profile/diarization are top roadmap item!
appreciate the pointer, will drop feedback on the github issue once I've put a few messier recordings through it rather than just a clean 1:1 call. good to hear diarization is an actual roadmap item and not an afterthought - that's usually the first thing that gets deprioritized under "good enough for now."
@galdayan Feel free to try it out! We think it works pretty great right now. But if you notice any room for improvement, let us know!