I tried 3 open-source alternatives to Otter for my meeting notes, and one runs entirely on my PC

Published Nov 11, 2026, 2:00 PM EST Abhinav pivoted from a career in banking to pursue his first love in writing. Even while working full-time, he continued contributing as an editor-at-large, a role he has held for more than 7 years. A lifelong tech enthusiast who has built three gaming and productivity powerhouse PCs since 2018, his passion for technology keeps him closely following the semiconductor industry, from NVIDIA and AMD to ARM. His MSc dissertation explored how artificial intelligence will reshape the future of work, reflecting his curiosity about the wider social impact of emerging technologies. Nearly everyone seems to have an AI-assisted workflow these days. There are many great use-cases for complementing one's workflow with AI tools, particularly those that reduce friction by automating repetitive tasks, organizing data to aid decision-making, and, of course, helping one recall things that they'd otherwise find easy to forget. The last one is perhaps the reason why I keep seeing everyone show up to Teams and Google Meet calls with an AI note taker like Otter sitting in as a participant. Although it's the most popular tool for the purpose right now, I've always wondered if it's a good fit for all workflows, particularly those involving sensitive information that an organization wouldn't trust a third party's servers with. To find a more local alternative, I tried three open-source tools, one of which keeps everything on my own hardware. With one scripted voice for zero ambiguity To give all tools a level playing field, I had every single one record the same Google Meet call on my PC simultaneously, with Meetily, Anarlog, and OpenWhispr, along with Otter running alongside for a thorough comparative analysis. The meeting itself was a three-minute scripted briefing from a fictional project lead named Sarah, read out loud by my MacBook Air's built-in text-to-speech function. I stuck to a single speaker on purpose, since Otter can tell Meet participants apart, and several voices coming from one laptop would've otherwise skewed my findings. I laced the script with 15 traps, including names, GPU model numbers, prices, dates, and specific deadlines for specific participants. For the summaries, I checked whether each tool caught both decisions and all three action items with the right owner and deadline. I also planted two hallucination traps relating to an undecided monitor upgrade, and a line stating nobody needed to copy files locally. OpenWhispr wrote the cleanest transcript But tidy isn't the same as faithful, as it seems OpenWhispr is an MIT-licensed dictation app that also records and summarizes meetings for users. Its setup offered on-device models, but I picked OpenWhispr Cloud since it needs no downloads and the free plan covers five hours of meetings per month. Its transcript was the tidiest of the three, scoring 10 out of 15, and only one point behind Otter. It was the only tool that heard the word "offline" correctly while others mistook it for "applying" or "lying." There was a fair share of errors, like naming the GPU "RTX 5070T", naming a version of Windows "Windows 1126H2" (which could've tripped up someone not familiar with them and led to an awkward Teams message later). What was slightly concerning was the fact that it swapped "the new 5070 Ti price" for "the new MSRP price", which changes the meaning completely. The summary still caught both decisions and all three action items with no hallucinations, though. Metric Score Transcript traps 10/15 Decisions caught 2/2 Action items (owner + deadline) 3/3 Summary hallucinations Where your audio goes OpenWhispr Cloud (On-device optional) Cost Free up to 5 hours/month Meetily kept the entire meeting on my PC It fumbled the names a bit, but not the tasks Just like OpenWhispr, Meetily's free Community Edition is also MIT-licensed, but with the key distinction that it does everything locally. It transcribes with Nvidia's Parakeet TDT 0.6B v3 (int8) and writes summaries with Qwen3.5 4B at Q4_K_M quantization, both of which sit on my drive. Contrary to my expectations, even with a respectable local AI stack underneath it, its transcript was the weakest of the bunch at 8 out of 15. There were several issues with the naming, but perhaps the most punishing to the reader would be changes to specific SKU names, like the RTX 4070 Ti Super becoming "470T Super". The summary repeated those naming errors, though it got the substance right, and that was the redeeming factor. It nailed both decisions and all three owners and deadlines, matching Otter squarely. Its only slip was filing the undecided monitor upgrade under "Key Decisions", though it did get it right further down. Metric Score Transcript traps 8/15 Decisions caught 2/2 Action items (owner + deadline) 3/3 Summary hallucinations 1 Where your audio goes Local Cost Free (Community Edition) Anarlog is local-first, just not on Microsoft Windows My audio left the PC, and some of it didn't make it back to the transcript Anarlog started as Hyprnote, and like the other two, is an MIT-licensed, bot-free note-taker that rather vocally pitches itself as local-first. On Windows 11, it's still in its beta, so that part of the promise is slightly misleading. It's also worth noting that built-in, on-device transcription is limited to Apple Silicon. Signing in dropped me into a seven-day Pro trial. That sends my audio to Anarlog's cloud, which currently routes it to Soniox, while Claude Sonnet writes the summaries. Once the trial ends, it's $14 a month against Otter.ai's $16.99 a month subscription. Anarlog was impressive as it was the only tool to get most of the product names right. It got the "RTX 5070 Ti" and "Windows 11 26H2", but the problem was that several words went missing all over its transcript. It had similar problems with participant names as the other open-source models, but what made it slightly worse was the fact that it entirely missed a participant's deadline. That put its score at 9 out of 15. Its summary still caught both decisions and 2 out of 3 action items, and instead of inventing a missing date, it admitted the transcript was garbled. Metric Score Transcript traps 9/15 Decisions caught 2/2 Action items (owner + deadline) 2/3 Summary hallucinations Where your audio goes Anarlog Cloud Cost $14 a month after 7-day trial Otter is still the one to beat, but it's no longer the only viable option Running this test made me realize why Otter is a default choice in so many calls. However, if you want the information on your meetings to never hit a cloud server for confidentiality or other legal reasons, you can't go too wrong with Meetily, which does that right out of the box. OpenWhispr is also a close second, provided you're ready to host a capable on-device model.