{"id":"47e45366-c240-4a25-b649-24fdc6d94f40","created_at":"2025-11-06T05:40:53.222192Z","started_at":"2025-11-06T05:40:55.264236Z","finished_at":"2025-11-06T05:55:05.990708Z","language":"en","source":"omi","status":"completed","visibility":"shared","structured":{"title":"Exploring the OmniDev Kit: Ethics and Open Source","overview":"The user shares their in-depth thoughts on the OmniDev kit, an AI-powered conversation recorder and transcription tool. The blog post covers the kit's features, integrating OpenAI for transcriptions and summaries, and the user's ethical guidelines for recording conversations. The user emphasizes strict boundaries for use: only recording where expected, in legal/risk situations, and for personal reflections. Discussion extends to the philosophical limits of AI—especially around summarization vs. creation—and the user's appreciation for the device's open-source nature, allowing local deployment and modification. They intend to experiment with self-hosted inference to maximize privacy and utility, motivated by concerns over data monetization by big tech. The post concludes by reflecting on how the device supports users with memory challenges and the potential for DIY customization.","emoji":"🎤","category":"technology","sections":[],"action_items":[],"events":[]},"transcript_segments":[{"id":"64604633-c8ae-4c06-be7a-bfa5733b35bb","text":"O I want you to take this verbal transcription and do very little to change or summarize it, period. Instead, I want you to take what I'm saying and turn it into a blog post. I'm making a blog post about my feelings about the OmniDev kit and I want to use the device itself to transpose this.","speaker":"SPEAKER_0","speaker_id":0,"is_user":false,"person_id":null,"start":0,"end":29.21},{"id":"47ab9b2e-e201-4c81-8abe-5fe11beb9670","text":"I want you to just output what I'm saying verbally into material that could be put on a page and a blog post. So I got the Omie dev kit and it's a tool that records conversations. That's it. However, the developers have added more. It integrates with OpenAI to provide this, like, transcription and summaries and action items and everything else. And most of this blog post is me ranking me for this kind of little fucking network and seeing how it looks like natively when transcribed by the device. I got it. I'm using it, and here's why. First off, I want to cover and hit the concept of recording people. Recording people is ethically a weird gray space filled with discomfort intrusiveness, and these conglomerations of what we've conceptualized for a private to be. But it feels icky to me to record people without telling them. So I define my boundaries for when I'm going to use this device. First off, meetings where recording is expected. Team meetings, client calls, interviews where everybody knows it's being recorded. The second point is where I am at legal or high risk situation. Any situation where I feel like my life safety or legality could be at risk, by the situation itself, it's turning on. But this is no different me recording on my phone. This is no different from me pulling out a pen recorder and just recording the situation. And then the third time where I'm using this is when I'm ranting when I'm talking, when I'm speaking at no 1 else but myself and to conglomerate my thoughts into something that I can look at later, have a high level overview. This includes like doing, to do tasks or pressing it, turning it on and saying, hey. Remind me to do this thing. And then it goes into my action items, which is really neat for me. So that's fundamentally how I feel about recording people And the next part I wanna talk about is AI. For me, a lot of people feel like AI is this weird like, single pane of glass. That fabricates and lies and hallucinates. And that's true. It does. Right? If you ask AI to create and generate, it's not gonna work. It just won't. Right? And for me, the usage of AI falls along the core tenets of translation because that's what the transform algorithm was designed to do. Transformers was assigned to do language to language translation. This can be also verbal to text or text to verbal. Speech to text, and text to speech. This is really, really nice. I use it all the time when I'm driving. And I get texts. I use my voice to text. To communicate because I don't want to text and drive. I don't wanna type. It takes away my concentration. But I can talk into it as if I was calling someone, and it doesn't take as much","speaker":"SPEAKER_0","speaker_id":0,"is_user":true,"person_id":null,"start":-0.42687499999999856,"end":268.12},{"id":"88d052b1-7d53-4fd5-adac-9fafeb7e1bac","text":"Some people do it better than others. Perplexity is a fairly good example of this. But fundamentally, that's not what AI was designed That's not what it's good at. That's not we've bastardized the transformers model and we've built things on it and we've added all like, additional aspects to make it work the way we want it to. But I don't think this is the infrastructure to do this less words to bigger words. Or the concept of creation itself.","speaker":"SPEAKER_0","speaker_id":0,"is_user":false,"person_id":null,"start":316.248289,"end":344.28828899999996},{"id":"afd90042-b3ce-41b7-bf68-646c2f028529","text":"I don't think AI is significantly when you start asking LLMs to generate and create things from inferences. When you go from less to more, when you go from little to a lot, But on the other hand, summarization and compacting is in a lot of ways destruction. And while it's time consuming, it can be done by anyone. So it's easy for AI to do this. But creation is a lot harder. These are more philosophical concepts. We could go deep into, like, how neural networks work and, like, why these things aren't working way we want them to work all the time. But that's not my intent here. My intent is to talk about this fun little device that is the first time I felt passionate about anything in AI in the AI space. Because I can see the goal and it's a doable goal. And it's something that creates actionable long term impact for people. It is nice. To have a device that records me when I'm in a legal, like, risky situation. It is nice to have advice that transcribes meetings that I have. I have a lot of fucking meetings, and it's kind of nice to have something that, like, shortens them down, gives me the high level overview, and reminds me of the things I said I would do so that I can be my word. And keep my integrity with the people in my life. And it's nice to have a tool that supports me in keeping an eye on things. That I forget. I have long COVID. And a lot of other weird mental stuff. And I have figured out ways to really build my life better despite that. The usage of calendars and tasking and creating support networks where people help each other and support each other in remembering things and creating things. And at the same time, there's still a small little gap. And I think this device can bridge that gap. In those moments where I might brain fog things out of my memory in 10 minutes because there's so much stuff going on. When I'm at a conference and I tell someone I'm gonna talk to them, and then I'm flitting away to the next person. It's nice to in the process of of flitting away, hit this button and say, hey. Remind me to this person from this company. And actually reach back out to them. And I think this is where the space excels. The other reason that this device and this is like sort of the next section here, the other reason this device excels is for me at least, the fact that it's so fucking open source. I don't mean open source in the way that, like, a of people are like, oh, yeah. We're open source, like, small amounts of it. No. This entire thing is open source. From the hardware to the firmware on the device to the app it itself to the back end infrastructure. They have put all of this on GitHub, and they have guides on their website. Showing you how to run your own back end inference server. They have guys on their website showing you how to rebuild the app. To use your thing running through Ngrok. And this is really, really interesting to me me because I just got my home lab set up with the help of a couple friends. Thank you, Trevor and Joan. And Asher. And I have a pretty good GPU. And so I wanna offload these, like, LLM tasks to my own systems because I don't trust organizations and companies that will profit off of the eventual of all the data that they're collecting. Because that's the long term goal of these Anthropic, etcetera, etcetera. A lot of it the way to make money advertising in this space. Google's biggest revenue comes from advertising. And OpenAI Anthropic and all these people are gonna start conglomerating data on people and start serving ads in specific ways. I believe this is gonna be true. We'll find out as time goes on.","speaker":"SPEAKER_0","speaker_id":0,"is_user":false,"person_id":null,"start":345.17828899999995,"end":617.0582890000001},{"id":"297ee64d-26ac-44ef-b387-860bddfb24e4","text":"But fundamentally,","speaker":"SPEAKER_1","speaker_id":1,"is_user":false,"person_id":null,"start":588.481434,"end":621.568309},{"id":"98883d27-467e-405b-9cbe-b26126162a23","text":"I want to be able to run everything locally. And this device the OMI developers, have done a really fantastic job doing this. So this blog post is sort of outlining my intent, outlining my feelings on this. And I'll probably start working. To run my own inference server. Right now, I'm using their entire service to, like, talk this all out. But I don't want to. I don't want to be limited to 1200 words or, sorry, 1200 minutes. I want to be able to run everything unlimited on my own devices. And the entire infrastructure of this device has been set up to do this. From the fact that they're running Nordic chip, and they've open sourced their firmware. And I don't think I need to modify the firmware on the device itself at all. All it does is records and sends it over. But on the app, it takes that recording. It takes your API keys, and it sends that information over to a service that translates your voice to text. A service that isolates your voice. A service that then takes all this information and summarizes it. Then a service that looks for all of that and, like, creates action items and all these other things. And I think I can run all of this locally. And I'm probably going to play around with this in the next couple of months. I hold no commitments on what I'm going to do next. But I wanna play around with this. I wanna see how this goes. I wanna see what success we have here. Because this is the most promising open source AI thing I've seen ever. They have made everything available to the consumer. To tweak and create and mess with. And normally, not an open source fanatic. I'm really not. I didn't even touch the concept of home labbing until I literally had 3 friends. Tell me exactly what to buy and then help me set it up. By running almost every command for me. I moved from Linux to macOS. Because things just were easier. I still use Linux consistently for my job on a daily basis, but it's containerized. It's I use Linux every day. But my system that I use at the foundation is a system that I can trust to just work. So this is an interesting journey into open source. I haven't messed with a project like this in all long time, and it might be as simple as me getting a back end server setup. Changing the app to communicate with that back end server, running it all through tailscale, replacing all the AI parts, with my own AI, local running models from Hugging Face I might call it a day. Or I might go further. I'm not sure yet, but this has been exciting to me.","speaker":"SPEAKER_0","speaker_id":0,"is_user":true,"person_id":null,"start":592.311414,"end":833.3082890000001}],"apps_results":[{"app_id":"01K89EAAY3XMJA0SJ7NSP6N5GT","content":"## Getting the OmniDev Kit \n- Got OmniDev kit for conversation recording and AI enhancements\n- Device integrates with OpenAI for transcription, summaries, action items\n- Using the device to test its native transcription by blogging directly\n\n## Recording Ethics and Usage\n- Concerned about ethical and privacy issues of recording others\n- Only uses device in meetings where recording is expected, high-risk/legal situations, and personal rants/thought organization\n- Uses for tasks and reminders, e.g., \"Remind me to do X\"\n\n## Thoughts on AI and Summarization\n- Believes AI is best for translation and summarization, not creative generation\n- Sees AI’s value in speech-to-text and summarizing/consolidating info\n- Finds meeting transcription and summarization especially helpful for memory and task tracking\n\n## Open Source and Local Control\n- OmniDev kit is fully open source: hardware, firmware, app, backend\n- Guides available for running own backend server\n- Wants to run everything locally due to privacy concerns with company-hosted AI services\n- Has a home lab and plans to use own GPU for LLM tasks\n\n## Practical Benefits and Personal Impact\n- Helpful for staying organized, managing memory lapses (e.g., long COVID)\n- Supports integrity by tracking commitments and reminders\n- Provides autonomy and privacy via local data handling\n\n## Next Steps\n- Plans to set up local inference server and run OmniDev kit fully locally\n- Will experiment with integrating own models and backend infrastructure\n- No firm commitments, but excited to explore further open source customization"}],"plugins_results":[{"plugin_id":"01K89EAAY3XMJA0SJ7NSP6N5GT","content":"## Getting the OmniDev Kit \n- Got OmniDev kit for conversation recording and AI enhancements\n- Device integrates with OpenAI for transcription, summaries, action items\n- Using the device to test its native transcription by blogging directly\n\n## Recording Ethics and Usage\n- Concerned about ethical and privacy issues of recording others\n- Only uses device in meetings where recording is expected, high-risk/legal situations, and personal rants/thought organization\n- Uses for tasks and reminders, e.g., \"Remind me to do X\"\n\n## Thoughts on AI and Summarization\n- Believes AI is best for translation and summarization, not creative generation\n- Sees AI’s value in speech-to-text and summarizing/consolidating info\n- Finds meeting transcription and summarization especially helpful for memory and task tracking\n\n## Open Source and Local Control\n- OmniDev kit is fully open source: hardware, firmware, app, backend\n- Guides available for running own backend server\n- Wants to run everything locally due to privacy concerns with company-hosted AI services\n- Has a home lab and plans to use own GPU for LLM tasks\n\n## Practical Benefits and Personal Impact\n- Helpful for staying organized, managing memory lapses (e.g., long COVID)\n- Supports integrity by tracking commitments and reminders\n- Provides autonomy and privacy via local data handling\n\n## Next Steps\n- Plans to set up local inference server and run OmniDev kit fully locally\n- Will experiment with integrating own models and backend infrastructure\n- No firm commitments, but excited to explore further open source customization"}],"people":[]}