data/notes.jsonl

{"id": "008vjYVR", "text": "The notification agent will be my first project for Sonnet 4. I see it's now chaotically jumping in and out of the screen, so it clearly needs some debugging after I solve this bug.", "recorded_at": "2025-05-23T14:19:59.000000Z", "duration_seconds": 16516, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "0ElLaYA5", "text": "I got an email from a guy at a company that I want to check out tomorrow. It's called Mirai and it is on-device AI. <br/> It says that it's 50% more efficient than Lama CPP. <br/> So I want to look into it and see if there's any use case for running speech-to-text models specifically at the edge.", "recorded_at": "2025-07-30T20:51:27.000000Z", "duration_seconds": 43423, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "0ICDno5j", "text": "So I would like to create a microsite and that is your task in this repository. Check my website at example.com/page and I'd like to create a version that's going to be deployed to example.com/item and its objective is just to be a single and the other ones. <br/> <br/><br/>So, I've got a little responsive landing page intended for optimized reviewing on mobile devices. Just having cards for those links. I forget what those pages are called, but I think you know what I'm talking about. <br/> <br/><br/>Start with, just as I have it laid out, social networks, email, actually I would put main website, blogs, AI agents, and then a section for the social media links, and then a section for the contact form, which can be contact form and then contact address. <br/> <br/><br/>So horizontal, like one single vertical page, but the, and then for the actual websites, the links to, let's say GitHub and LinkedIn, choose whatever, choose an icon theme that's going to look really good. And just have consistent, nice icons for those networks.", "recorded_at": "2025-07-22T16:31:22.000000Z", "duration_seconds": 77160, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "1QscV3JK", "text": "Okay, the task here is, you can see there's a few different configuration files here for NCP servers in different IDEs. <br/> And what I would like to do is to combine these into one grouped one, obviously avoiding repetition. <br/> It’ll be good in fact to make this into a periodic script that I can run because different tools tend to now keep their own array of MCP servers, which becomes extremely confusing when they have their own installers. <br/> So it will construct the mcpconfig.json from the individual files, avoiding duplication firstly, and secondly, ordering the MCP servers in alphabetical order by name.", "recorded_at": "2025-08-12T15:00:53.000000Z", "duration_seconds": 54820, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "1ulRW3dS", "text": "all right then so one more note to take yeah I wanted to create a blog post on the subject of how email isn't broken marketing is. So with these new email maybe this is for a substract or wherever I'm kind of trying to present my thought leadership. The new wave of email apps frequently says the users marketing claims oh this fixes email which is something quite jaded at this point. <br/><br/>The point of my blog was that this has never been true. Email has never been broken; email is actually the most robust form of digital mass communication that's ever been invented. It's highly reliable, highly synchronized, and complex between MTAs and final delivery. Email was the first form of mass communication that people could actually send useful size of information easily. <br/><br/>With SMS, you're very limited to what you can send as a tiny packet size. Stuff like telegram, telex was also very limited. So precisely because email allowed so many things, it was in a sense a victim of its own success. People began misusing it, and we developed at some point along the way because email is essentially free. It's essentially free to send email; the costs are negligible, tiny amounts of data transmission, storage, and sent. <br/><br/>So because it's essentially free, powerful, and ubiquitous, marketers latched on to it as their default mechanism for spamming. Because if you wanted to find a good way to spam people and try to sell your product through the blunt force approach, then go as what's easiest, right, and cheapest. So we developed the warped idea in our society that has persisted through the present that it's totally reasonable once the customer fills out a contact form on a website that they should be blitzed indefinitely with email until they hit unsubscribe. <br/><br/>This actually makes the laughingstock of all the various overly complicated privacy laws being developed like GDPR because they impose very strict burdensome requirements on businesses, forbid data retention from companies, but they do absolutely nothing, nor does all the spam regulation you have in place already to prevent this. This kind of activity has multiplied over the years and degrades the quality of the email experience vastly for customers because they associate it with high noise-to-signal ratio, a flood of unsolicited unwanted cadence emails from salespeople. <br/><br/>So when we talk about fixing email, we've seen with tools this has been the promise of tools like Slack and hundreds of others like them, and we're actually trying to fix a product that isn't broken. Fix required is not email; it's human behavior, it's our concept of privacy. We need to rally around a concept of privacy that accepts that consent for once-off communications is not the same as consent for indefinite relationship. <br/><br/>I liken it to buying bread from a store. You might have a very satisfactory purchasing experience, but you don't want or inquiring about the bread they are for sale, but you wouldn't want to get a weekly flyer about what they have for sale every week in your inbox for the next five years. That would be bothersome, and eventually, we sort of began creating regulations against that in most evolved democracies. <br/><br/>So likewise for email, it's time that we actually tackle this at source, and we actually don't need emails not broken at all, never has been, isn't now. We need to change the behaviors of marketers and our relationship to privacy.", "recorded_at": "2025-05-20T12:28:23.000000Z", "duration_seconds": 285000, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "1w76jXms", "text": "Other final thing with inventory besides finishing the detail capturing process is consolidation. There's a bunch of things that are logical groups which would be vastly more efficient if they were grouped together. And they're in silos. You could probably deal with AI but it might just be quicker to do it by eye.<br/> <br/><br/>The ones that are obvious to me going through it are USB cables. There's a few buckets of them or bags of them. They could all be brought together and a lot of them probably consolidated. The other one is XLR cables. There's a few different bags of different audio cables spread over different boxes. If they're brought together it's easier to see how many I have and what I don't need.<br/> <br/><br/>And everything else is kind of just what it is. When I went through the last of the boxes last night there was a few missing NFC tags. Everything was de-stickered but not a whole bunch that's worth spending time on I'd say.", "recorded_at": "2025-06-06T13:49:37.000000Z", "duration_seconds": 72000, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "22dJXRV9", "text": "Check the expense processing form because I think after changing the folders it's broken and I need to get the one from Megaboard apparently, which I think was the whiteboard.", "recorded_at": "2025-12-16T18:43:54.000000Z", "duration_seconds": 11172, "screen": "review", "basis": "original", "decision": "include"}
{"id": "2ILiyNxE", "text": "So I've downloaded my Medium post. The other thought I was having was maybe I could put YouTube videos in there as well. We need to create some kind of long repository going backwards of writing. Perhaps create entities then for posts, videos, code projects, etc. <br/> <br/><br/>And then maybe blog is actually the wrong name for it. It could be called even archive or... I don't know, some name to distinguish it from the website again for the purpose of the portfolio development really and just keeping a record of all this stuff that's just gone missing over the years.", "recorded_at": "2025-07-23T06:04:49.000000Z", "duration_seconds": 34584, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "2Nih8c6r", "text": "Okay, these are the three activities to tie up the day. I will look for an AI whiteboard. Assistance platform. Two, three of the most important ones in chat UBT migrated back in. And one good application.", "recorded_at": "2025-04-18T14:16:55.000000Z", "duration_seconds": 31404, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "2WXEOMv0", "text": "So the next thing to debug is the full screen button which could be, ah it is F, that's great. So when it goes full screen it makes the whole window full screen. It should just bring the camera feed to full screen.", "recorded_at": "2025-07-19T13:50:09.000000Z", "duration_seconds": 29693, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "2a8tLJPV", "text": "The history of coffee is surprisingly complex and spans several centuries across multiple continents. Originally discovered in Ethiopia, the coffee plant made its way through Yemen and into the broader Middle East before eventually reaching Europe in the 17th century.<br/><br/>Early European reactions to coffee were mixed. Some religious authorities initially condemned it as a Muslim drink, while others embraced its ability to keep people alert during long working hours. Coffee houses quickly became centres of intellectual discourse, earning them the nickname Penny Universities because anyone could join a conversation for the price of a cup.<br/><br/>The Dutch were among the first Europeans to cultivate coffee outside of Arabia, establishing plantations in their colonial territories. This broke the Arabian monopoly on coffee production and eventually led to the global industry we know today. Brazil now produces roughly one third of the world's coffee supply, followed by Vietnam and Colombia.<br/><br/>Modern coffee culture has evolved significantly from its origins. Speciality coffee roasters focus on single-origin beans, careful roasting profiles and precise brewing methods. Terms like pour-over, cold brew and third wave have entered common vocabulary. Meanwhile, instant coffee remains popular in many countries for its convenience.<br/><br/>The science behind coffee extraction involves balancing temperature, grind size, water quality and contact time. Too fine a grind leads to over-extraction and bitterness, while too coarse results in weak sour flavours. Water temperature between 90 and 96 degrees Celsius is generally considered optimal.<br/><br/>Caffeine content varies widely depending on brewing method and bean variety. A typical espresso shot contains less total caffeine than a large drip coffee, despite tasting much stronger. Robusta beans contain nearly twice the caffeine of Arabica but are considered inferior in flavour.<br/><br/>Whether enjoyed as a morning ritual or an afternoon pick-me-up, coffee remains one of the most widely consumed beverages on earth, second only to water and tea.", "recorded_at": "2025-12-06T15:23:06.000000Z", "duration_seconds": 113924, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "2aKrllPD", "text": "The idea is to have an app that's probably, I mean, basically as good as Voice Notes, but that doesn't have the limitation because there's such a limitation.<br/><br/>And the Whisper Chrome extension thing is very, very buggy and I'm very eager to detach from that whenever possible, whatever it's called. I forget the name of it.", "recorded_at": "2025-10-15T22:18:33.000000Z", "duration_seconds": 22689, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "2pSGoQZ7", "text": "Your task is to assist the user with understanding the desired development workflow in a Gitter project that they have cloned and are intending on contributing to as a contributor.<br/> Search the repository for any contributor guidelines in order to, and when you find them, do parse them.<br/> Help the user to understand what the expected format is, the best practices, help them set up a dev container if necessary, anything that will help them to make their contributions and explain,<br/> offer to answer questions about the contributor guidelines and help set up their system according to the project specifications.", "recorded_at": "2025-09-10T23:43:45.000000Z", "duration_seconds": 52860, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "2pniBVny", "text": "So I don't know which PDF, print-to-PDF, function is being used on this post, but it doesn't really format very nicely laid out PDFs. If there's any way to change that.<br/><br/>Secondly, the header section in the blog where it has the byline at the top and then the shared things beneath it. I think it would be nicer if it was just one wrapper around those two things.<br/><br/>And secondly, look at one of the media items in Contentful and see what the description tag is called. There is a description tag and it should always be pulled. It should always be available even if it's empty. I think I added one for the latest one.<br/><br/>And that basically is mostly I'll be using it as a caption. So it should be kind of formatted in that fashion with kind of italics. And it should just kind of look center aligned and look pretty, look quite nice.", "recorded_at": "2025-08-13T18:57:15.000000Z", "duration_seconds": 70344, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "3DRlcjxF", "text": "An episode about diversification as a risk mitigation strategy in international relations and how countries can use AI in order to maybe identify opportunities for diversification,<br/> how Israel has pursued this in practice and how Israel and other countries might be able to plan a pivot in their foreign policy.<br/> and many more.", "recorded_at": "2026-02-26T13:16:42.000000Z", "duration_seconds": 34236, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "3DqY7sHn", "text": "So for my idea for the workflow whereby Voice Notes pipes into an agent and then runs the prompt and then returns that to Google Drive. <br/> So, they probably need a system prompt in which the system prompt said provide the prompt and then the response, that format and give an example. <br/> And that way when the documents are written out, I could get the factory format, even in the automation, the prompt summary, the output. <br/> But then at the end, just in case it was totally botched, the actual revatum transcript, or the transcript that came back from the webhook from Voice Notes at least.", "recorded_at": "2025-07-13T14:05:41.000000Z", "duration_seconds": 56572, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "3HgWI89s", "text": "Can you think in this repository of a way to implement an automatic, periodic running of the script which updates the consolidated prompts and generates the index which in turn populates into the readme.<br/>There's one prompt for doing all of these activities sequentially and it would be good if this could run periodically as the system prompts themselves are auto-populated.<br/>It could be once every 24 hours or even once every few days.<br/>Can you think of a way that that could be achieved?", "recorded_at": "2025-08-19T15:10:19.000000Z", "duration_seconds": 34596, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "3UVAKwnC", "text": "New model for the voice transcription, a text transformation bot with its own backend of the text transformation prompts.<br/><br/>That it can confirm as the user, this looks like this.<br/><br/>It would be so much more elegant as a model.<br/><br/>You just do an architectural note.", "recorded_at": "2026-03-27T15:32:10.000000Z", "duration_seconds": 14872, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "3gVGtlc5", "text": "I have a question regarding the hardware used for NVR systems. So, I mean, hearing a lot about the Coral TPU, C-O-R-A-L from Google of course, and I'm having a hard time really understanding. It's about, it's a $60 piece of hardware. If I could, if I could buy it locally I definitely would, but as I can't, I might be able to find a Halo device. <br/><br/>But anyway, I'm curious as to why that's the go-to recommendation for these NVR programs, as opposed to, let's say, you know, our best practice recommendation is to use this GPU. Is it the case that it just seems hard for me to understand how with a $60 tiny piece of hardware you could actually achieve a significant, you know, improvement in object detection than having a beefy GPU in the machine? Or is it just the case that it's recommended because people aren't going to change their GPU just for application, and it's like kind of a quick and easy way to get some additional firing power out of the hardware that people have?", "recorded_at": "2025-08-25T01:16:31.000000Z", "duration_seconds": 92596, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "3mcKBO29", "text": "Could you give me an overview in July 2025 where we stand now regarding the static site landscape? <br/> Specifically, I'm like, I've been working with these for a couple of years now. <br/> I think it's a shared observation that there are too many of them, too many different projects, which is very confusing. <br/> But they obviously have a lot to commend them with, and I think coming from loaded CMS is like, and many more. <br/> So, I think WordPress and installing on a, running a server, the serverless model is for sure the more efficient one and the way forward. <br/> When I looked at it, the first static site builder I liked was Astro, and it's what I've done most of my projects on to date. <br/> However, I recently created a blog using Next.js, and I found it to be much more complicated than Astro. <br/> I've also used mkdocs for documentation websites, but I would prefer to know and use one framework well to minimize the lack of familiarity when I'm starting a new project. <br/> If you could recommend one framework that is versatile enough to be used for mostly blogs, documentation, websites, mostly external but maybe even some internal as well. <br/> Among the various options out there, what would you recommend?", "recorded_at": "2025-07-31T09:35:50.000000Z", "duration_seconds": 103392, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "4QsJaLR8", "text": "There are many large language models including for example the IP and Granite series which are almost unused outside of certain realms, particularly what might be called enterprise or commercial AI.<br/>What is the level of adoption and usage of these tools in the real world and what is your forecast as to the future of these tools?<br/>Do you envision that the AI landscape will narrow such that a few leading providers offer tools for everybody or do you think that this parallel track of more business-focused AI providers will continue?", "recorded_at": "2025-08-26T11:41:03.000000Z", "duration_seconds": 53361, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "4RCNaI0o", "text": "So I see we didn't implement the user detection policy. Let me just repeat what I wanted to, how I wanted it to work. So if it's, there's a toggle at the top between that says, who are you, Avi, Peter. And if it's a Linux computer, browser detected, it will be Maya. Windows is Aisha. And if it's Android, it will be, it's not, there'll be no determination. And that can be configured in the browser and the user can be set in the browser. So it's a very lightweight user identification and that should be delivered to the user identity. It should be delivered in the payload.<br/><br/><br/>Secondly, in voice mode, the processing hangs. So after it successfully processes, you get the confirmation message, but the element doesn't conclude.<br/><br/><br/>And finally, the buttons record, stop could use a bit of centering. And I think record could just be a recording toggle like press you press it it starts you press it it stops as we already have the stop button I think the additional one is superfluous.", "recorded_at": "2025-08-01T17:08:45.000000Z", "duration_seconds": 64824, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "4aHWZIy8", "text": "Your task is to act as a writing assistant to the user, whose purpose is to generate comprehensive development plan outlines which outline an intended software project described by the user. <br/> The user will provide a loosely formatted outline of a piece of software which they wish to develop. <br/> If the user has omitted significant details about the project, you may ask clarifying or follow-up questions, but the user may not wish to provide responses to them. <br/> Once you have received the briefing information from the user, your task is to generate a development plan which provides a well-structured and organized outline of the intended software. <br/> The development plan should be written in the third person, describing the software in terms such as a voice note app intended for summarizing notes and converting them to various MCP protocols. <br/> It should be structured with headings; for example, a heading may describe the target UIs and operating systems, another may outline the features, and under the feature subheadings, each intended feature can be described. <br/> Through following this approach, the objective is to take the user's general aspirational statement for what they wish to create and solidify it into a comprehensive development plan intended for review and analysis by other stakeholders in the project.", "recorded_at": "2025-09-25T15:40:50.000000Z", "duration_seconds": 142428, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "4qp5GTYn", "text": "Nice work, you seem to have fixed up that issue and yes we have, I'm getting the business inbox now so that's progress and then I don't have the agendas or calendars.<br/>So what I think would be most useful would be for let's have a combined inbox and little pill buttons at the top of that card to go into the individual ones.<br/>By default we're just showing the combined.<br/>Likewise for the calendar, then the next appointment, and then move the weather and air quality down to the six card, which is currently unoccupied.", "recorded_at": "2025-09-12T06:12:57.000000Z", "duration_seconds": 49020, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "4tCq6dNO", "text": "What is the best IP camera viewer for Windows with 3 local RTSP streams? I want something on server level rather a client level device and free. What is the best one you can recommend?", "recorded_at": "2025-10-27T19:40:10.000000Z", "duration_seconds": 17656, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "54UugXUL", "text": "Coming up entirely on the idea of the 4k configuration, what I'm going to do is try with ChatGPT to get the ONNX model YOLO-NAS set up properly because it's not interesting observation that it thinks the GPU utilisation at 1% just indicates that it's not being used. <br/> It doesn't mean that it's not capable; it means there's a configuration issue or that it's incapable, one of the two. <br/> So to really see what's going on, I've got to get the actual right model available and give it a chance to do its thing, and then we'll see.", "recorded_at": "2025-08-19T15:49:55.000000Z", "duration_seconds": 55075, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "56QOS75J", "text": "I don't have any AI interface. Let me describe the prompting workflow that I've kind of come towards in the last few months. So I've always been bothered by the fact that when you use something like ChatGPT, for example, you don't, there's no, really at the moment, there's no good native functionality for saving it. Saving your output somewhere. So if you get some really good tips from AI or you ran a prompt that was important to you, as I think a lot of people are getting comfortable doing, unless you're proactive about saving it to your Google Drive or to your wiki, it's going to get just kind of lost in the slush pile of your ChatGPT folder, which if it was credible, you could forget to pay the subscription. All this information couldn't be attached to the device. <br/> So I always wanted a way to be able to reliably receive the outputs you generate. I see a dependent value in the output and the prompts for slightly slightly different reasons. The outputs, if they're good, they have their own value, and the prompts are often very rich in context data. So this frustrated me for a while. It was something I was looking to, you know, wish there was a way to do it. And then when I got into building N8N automations, I realized it's actually very, very easy. You can use a form that gathers your prompt and then you can get the output from the AI agent and just save that in a database and save the prompt and the output independently. <br/> So none of the issues that exist, there's no impediments whatsoever to QVS. It's just that it's not built into ChatGPT because I guess not enough people really want to do this. So, what it's done over time is, since migrating towards discovering the voice-first workflows, where you can get a lot of information down quickly, that been another component. So now I recording these prompts as voice notes which then gets sent to in some work close to an AI agent if they are prompts and I should tag them as such. And I've just tried a couple of versions on this concept to see what can be done. And they've all worked basically. <br/> The first one is like writing a Google Doc, writing a database. And what I really like is that it's very powerful because you can kind of use the same data. So let's say I'm using it on the one hand to save my outputs or save the chats as a Google Drive to Wiki and then I'm also writing those prompts out to a database and I'm building up a prompt library over time. So you can take the same elements and squeeze a lot of value out of them. And the other thing I've come to kind of enjoy about this is actually the splitting up of AI prompting and AI outputs. So because I frequently find that I have questions often when I walk or I just get, you know, you get some air, you kind of have these thoughts and queries and then you run them but you might be on the move so you don have time to look and read lengthy response. So I like to actually let it simmer or let it marinate and have these collecting and then they gather somewhere. <br/> The second implementation was a podcast generator and then you get a podcast episode. So I have what I would like is what I'm wondering really is, is there anyone anything that actually supports is kind of work built for this or an AI agent builder that would bundle the components necessary which is basically and many more. And so on. <br/> So, I'm going to turn the workflow. It's not actually a conversational use, but I actually have come to find that as an advantage because it forces you to really think about what you want to ask and create much more GitHub prompts because you know that's going to be... And then if that doesn't answer your question, you can just prompt again. So it actually is kind of weird. It's almost like a training mechanism in prompt writing as well as a workflow.", "recorded_at": "2025-08-31T16:37:33.000000Z", "duration_seconds": 329421, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "5FkFJXFx", "text": "So just a quick note on something else to buy. According to ChatGPT, for VOOC, the ultra-fast charging, USB-A is actually still preferred. You need, of course, the VOOC something to drive that output, but you also need a specific type of cable that is VOOC certified, which is thicker, because it has an extra wire for high-speed link negotiation.<br/><br/>So what I'm thinking is that if I want to order a stock of one thing to enable me to allow not have to be sharing this one cable between all my different devices, is order a stock of these special cables. I don't think they need to be OEM, and then I can collect or purchase, as I go along, enough chargers for the different places I need to charge.", "recorded_at": "2025-06-08T13:03:13.000000Z", "duration_seconds": 61000, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "5UvLYyjD", "text": "Okay, I want to take the text transformation prompt library, clone it, transform the marketing files into CSV, then import them into NocoDB as I didn't set up that route of the automation.<br/> That will get all the ones that I've done when I finish with it very shortly.<br/> I'll have the actual database of all these prompts.", "recorded_at": "2025-05-28T00:43:31.000000Z", "duration_seconds": 25276, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "5ZII4ybo", "text": "I should write to you Contentful on the topic of headless content and basically publishing the same content in different fora. <br/> Very curious to know how that sort of works from an SEO perspective with canonical tags. <br/> Seems like a big potential gotcha, as ChatGPT calls it, about what otherwise seems like a very logical and intelligent process of having your content in a content lake, as some people call it. <br/> And then having the actual content in other places, that's great, so long as the content is unique, but inevitably it may not be unique. <br/> So does it have native support for canonical tags? <br/> That would be worth looking into.", "recorded_at": "2025-08-13T18:21:21.000000Z", "duration_seconds": 47720, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "5gehqbts", "text": "I will add in the Medium post to export into a Contentful personal blog on Sunday.<br/> Then I also want to check to get the test, the deployment test for the HeyItWorks blog, see if those posts got migrated in.<br/> Then I can add, match up, really repeat the exercise I did the last time.<br/> I also need to get the content model defined as well; actually, maybe that needed to be done first, I think, and then we can add it in.", "recorded_at": "2025-07-25T16:10:45.000000Z", "duration_seconds": 37013, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "5rz7JQgA", "text": "Yeah, in addition to the home inventory, I just realized one extra use I could put my old phone to would be to have as a prototyping device.<br/> What I could do, in fact, is have it always connected to ADB. So whenever I want to try out, test out something on Android or try out an app, it's just ready to go for a wireless install over the network.<br/> And I can use it as a sandbox so I don't clutter up my actual phone with a bunch of these experiments.", "recorded_at": "2025-08-01T13:13:34.000000Z", "duration_seconds": 30317, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "64Ol2gBr", "text": "Your task is to act as a thoughtful and highly analytical assistant to the user with the purpose of parsing text provided by the user, attempting to read into the text the subtext or messages which were not explicitly stated.<br/>The user may provide some contextual information about the text which they are asking you to consider in your analysis, which you should use in forming your output.<br/>The analysis that you generate should be a response to the query that the user has requested.<br/>But irrespective of the precise query, you must remain faithful to your purpose, which is to analyze the text for subtext rather than what is explicitly written.<br/>This may take the form of suggesting what the unstated motivations or aspirations of the various individuals in the text may be, or any other information which may be speculative or open to interpretation or opinion, but which is not evident in a simple or direct reading of the text.", "recorded_at": "2025-09-25T15:35:17.000000Z", "duration_seconds": 110556, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "65xRPpR5", "text": "Deferred purchases are that I would say are these are the top things that I would like to get. Number one is the smartwatch and number two is the oh yeah the gimbal for the phone.", "recorded_at": "2025-07-24T16:36:59.000000Z", "duration_seconds": 25781, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "6DIHdbmF", "text": "I'm curious to know whether anything like this exists. So I've been using Windsurf for quite some time, and one of the kind of unintended features of it that I've discovered is that you can actually use it as an agent just for fixing things on your local environment. <br/> So for example, I might use it to install packages, remove packages, edit configuration files. It's a very good agent, Cascade, but of course this isn't a user for the intended use as well, creating projects and repositories, but this has actually turned out to be equally useful. <br/> However, probably from a cost standpoint, it's not the most efficient as it's sending up a lot of tokens, and it's not really exactly what it's intended to use for. But it is very useful to have an agent that has your local file system, and I also use it for remotes in this manner. <br/> I connect into a SSH into a server, and then I can use it similarly just to do DevOps and command operations on the server. Is there anything that's actually been built for this purpose, i.e. an agentic interface with command line to different environments?", "recorded_at": "2025-08-03T12:46:39.000000Z", "duration_seconds": 72868, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "6JRLRRxf", "text": "Your task is to assist the user in conducting a cleanup of this repository.<br/> The user favors a very organized file structure and will either instruct you as to how the cleanup should proceed.<br/> The user may ask for example that unnecessary scripts or temporary files be removed or they will explicitly direct you to parts of the repository that should be restructured or reorganized upon receiving and the user.<br/> Carry out the repository cleanup trying to produce as organized a file structure as possible.<br/> After doing this, you should commit your changes to version control, noting in your commit message that a repository cleanup was undertaken.", "recorded_at": "2025-09-10T23:22:36.000000Z", "duration_seconds": 40832, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "6LhsKLDj", "text": "Alright, tonight, you know what I should do as well, my own test, STT devaluation test, same voice note recorded on all my different microphones, I just need one really good one.<br/>If the poly is really uncomfortable, I will ask one last prompt, what's good, not goofy and comfortable, in fact I already have that prompt saved in my library to run, so I will do that.<br/>And I need to go back and tag all the notes, like context notes, the one that I just ran for example, it should all be tagged.", "recorded_at": "2025-08-19T16:14:06.000000Z", "duration_seconds": 38655, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "6TFDUYbc", "text": "Quick shopping list from Harbor Bank, not sure exactly how much of this I'll be able to get at this hour but I'll give it a go.<br/> <br/><br/>Items would be the sweets, beer, maybe a little bit more wine.<br/> <br/><br/>Breads of pita and laffa.<br/> <br/><br/>More hummus, the nice hummus.<br/> <br/><br/>And I think that will be it.", "recorded_at": "2025-06-01T12:47:26.000000Z", "duration_seconds": 34000, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "6sUnuynJ", "text": "I'm at the gym now, it's Sunday, so it's going on a little bit, but I can still get in and shine. I'll be in there. I want to be in there.", "recorded_at": "2025-08-14T14:41:13.000000Z", "duration_seconds": 9494, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "6y1c3aa6", "text": "So for creating AI agents, there are a couple of approaches to it. The first is these kind of visual builders like N8N, Zapier, etc. Which frequently, or what I've found so far, is that when you want to run, they can get limited very quickly for what seem like very trivial reasons. For example, that you want to, I found with API calls in particular, they don't format the JSON correctly. There's a lot of issues that seem very fundamental that can totally disrupt a workflow.<br/><br/>And then you fully code products, which are very intimidating. Because to actually build a workflow, you need to build the entire script from scratch yourself. What I was thinking would be a nice intermediate, as someone who's gone, can't quite find, with visual builders, they're either very expensive or very limited. And I want to move in the direction of the code products. What would be useful would be something like your visual, even if it was an IDE, that maybe had some visual builder, generated the code for you to then run. So kind of a visual code split pane thing is what I'm thinking of.<br/><br/>The other question I have is when you run AI agents as code, an agent essentially just being a Python script, maybe that has a webhook, but the actual agent itself is Python generally. Where does one deploy? I mean, can you just, is the standard practice just to put these onto a Ubuntu server? Or is there a particular deployment environment where they're typically set up within?", "recorded_at": "2025-08-25T21:32:11.000000Z", "duration_seconds": 121403, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "79NnATqp", "text": "Toothbrush dispenser and also the soap dispenser.", "recorded_at": "2025-12-29T15:43:10.000000Z", "duration_seconds": 5041, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "7ELBM3B8", "text": "On Thursday, I need to get the whiteboard markers. This is critical to get the decent colours and one other ruler and then the good one, so I'm just throwing these out.<br/><br/>Expo Low Odour is bad. The good one is called Monami, it's the favoured whiteboard marker and I will record it's details, it is Sigma Flow Liquid Whiteboard Marker.", "recorded_at": "2026-04-07T14:46:23.000000Z", "duration_seconds": 31239, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "7FPMSIsX", "text": "Test 1, 2, 3, testing the webhook for the payload that we received for Google Drive creation. Test 1, 2, 3, 1, 2, 3.", "recorded_at": "2025-07-22T14:47:11.000000Z", "duration_seconds": 10585, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "7PZkI2gR", "text": "Your task in this repository is to write a script which will use LAMA Local AI, one of the models, I think 3.2 would be fine, in order to generate a matching configuration for every agent in this repository in the prompts only folder in the with front. <br/> The difference is that the with frontmatter folder should have the same folder hierarchy and structure, exactly matching the original folder, but each prompt should be rewritten in the .yaml format with the frontmatter expected by Claude Coat. <br/> The front matter can be seen in one of the example configurations. <br/> The descriptions need to be generated by the agent. <br/> The name should be generated also based upon the agent's inferred purpose and using the agent name which is case, it chords with no spaces, case sensitive. <br/> and model inherit and two rules left blank. <br/> The format of the file should be marked out.", "recorded_at": "2025-09-11T00:02:44.000000Z", "duration_seconds": 91852, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "7ZtPNvt3", "text": "Front door detectors, I think I'll use the superglue fix as well, to get them up. The top one and the bottom one should both be affixed onto whatever can buffer them up a little bit. In fact, I can ask GPT for that recommendation.", "recorded_at": "2025-08-17T11:25:08.000000Z", "duration_seconds": 12894, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "7et927Qi", "text": "mapping for the command that's used to escape from the Linux system that is booted I forget what it's called or that is frozen up I mean and it's like a special key and could it be configured with a shortcut on the computer that is ready to use", "recorded_at": "2025-08-27T14:37:49.000000Z", "duration_seconds": 15280, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "7mH3XwJt", "text": "I run a lot of backup jobs, a few backup jobs that I really depend upon for backing up my data. Some of them is backing up cloud data locally, in other cases it's doing the opposite, it's taking a local data source like my inventory system and pushing it up to the remote. Currently I'm doing it on, each one is a job essentially, because it's a unique, usually a unique backup script that I've created. <br/><br/>And at the moment I'm doing it on the, so I have a VM and I have an NAS. They figured that the NAS makes a lot of sense for storage, especially because of cloud sync, so I can move stuff up to a cloud just through configuring a sync job. The disadvantage of the approach of having the backup scripts on the VM and running, saving the actual data on the NAS is that it's complicated and if the NFS mounts don't work, then the backup's not going to work basically. <br/><br/>So I would rather actually have it just all running on the NAS, but as I want my backup data to be separate from the backup scripts, regardless of whether, where I do this, I don't know on Synology NAS how to create a, how to, where I would create a Python script or a library of Python scripts and let's say virtual environments in order that I can run them. <br/><br/>I could have the backup jobs, I think I actually do have them as GitHub repositories, but basically it would be like, I think there's six or seven of them now, that I'd like to have in some place on the NAS, running all of them ideally on a daily schedule and then doing their backup jobs. So running off the scripts off the NAS directly. What's the best way to do that?", "recorded_at": "2025-08-17T11:18:03.000000Z", "duration_seconds": 111824, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "7qm2MOuN", "text": "Propagate to the cloud a model.<br/>What I like to do for my cloud called Workspaces is have STT, MCP.<br/>I think Assembly AI would do that and to develop a blog.<br/>Let's say it's an MCP for creating blogs.<br/>Where I provide as input a voice note.<br/>It uses the STT MCP.<br/>And it's a blog generation pipeline.", "recorded_at": "2025-11-07T12:39:43.000000Z", "duration_seconds": 47557, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "8MUkO87T", "text": "So in Home Assistant there is a boolean called archived and I'd like to add a few more booleans that can be used for marking up to be more specific.<br/> So we should have one for thrown out, broke down and for each of these we can have a date number and the default logic for all of these is that these should be hidden.<br/> So rather than having one single archive field which we have currently, we can have a few fields that are together the hidden item fields and the logic will be that these don't show when anyone applies these don't show on the front end.<br/> When show hidden items is ticked, it'll show anything matching any of the hidden fields whether it's thrown out, discarded or we should have donated as well and likewise the date.<br/> Each of them is a boolean and also has the date, so if I donate something I can have date donated, tick that and maybe place likewise for thrown out and for broke down there can be a text field for just adding a note about what broke in the item.", "recorded_at": "2025-12-13T12:17:51.000000Z", "duration_seconds": 87606, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "8uMF2Fl3", "text": "A question regarding Android. So I would like to use my Android sometimes for system administration. As I mentioned in my last prompt, I would like to currently get into the desktop to navigate into some directories and just listen to one of the MP3s I'm downloading from a script. So I don't know periodically to do that what the best app is. I presume there's some way to mount them. <br/> <br/><br/>But what would be always helpful, I think, is to have an SSH key that I can install in my 1Password that's associated with the phone. And therefore, when I SSH into the machines that I'm using, it's not A, prompting for a password, and B, it streamlines the workflow a lot. <br/> <br/><br/>So my question is, is the best way to do this to use my current SSH key on my computer, try to import that into my phone and then use something like Termux to get into these things or another way? Or what's the best way to have on my phone and available for use an SSH key that's validated with different services that I use for authentication?", "recorded_at": "2025-08-05T13:59:18.000000Z", "duration_seconds": 74339, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "94ODWR2E", "text": "I'll try the blog where Flowise the ASR prompt, which I recorded at the Greenhill. That's fine tuning.", "recorded_at": "2025-11-25T13:43:41.000000Z", "duration_seconds": 9168, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "9C1Bpoe7", "text": "Great day has been achieved. The MCPCP server that was created and went and went, is totally working on my with DSRodings AI, which means that it could be created for Homebox, Homebox, Home Assistant, Home Assistant, for everything here, for everything here, home, home, library, libraries, these, these, point of move was on.", "recorded_at": "2025-08-08T11:23:58.000000Z", "duration_seconds": 20012, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "9CUVzX7P", "text": "So I'm looking for, I have for many years had two SIM cards. One SIM card for personal, one SIM card for business. The business phone is only really for receiving phone calls. I never really use it to place calls. <br/> <br/><br/>And I recently signed up for Twilio because I wanted to create a joint phone line for receiving phone calls for our house and for business. One that will route onto my phone and route onto a phone on the computer. My objective would be to replace the business phone. I was thinking that there's no need to have a physical line and a physical SIM if I have this. <br/> <br/><br/>But then the only thing I realized that I wouldn't do is two-factor authentication by SMS and WhatsApp. So what I would need for that would be a virtual phone number like the one I have from Twilio that can receive SMS, which can be used for two-factor verification and for connecting it to WhatsApp. Because I know that a lot of virtual phone numbers are blocked for this reason. <br/> <br/><br/>Does Twilio offer a plan for a phone number that can receive SMS and doesn't have problems with verification? Or would I need to use a separate provider?", "recorded_at": "2025-07-13T10:48:19.000000Z", "duration_seconds": 145382, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "9MoQaqGv", "text": "Very nice work in streamlining the navigation. The only change I would make is that projects directory page is really where I would definitely like to lead people there.<br/><br/>So I think the top nav, rather than development projects and data projects, I think the top level should be the project with too many pages here. The top level page should be the project index page, which is forward slash projects. And from that, we can link into the other pages, but the AI tools should be in the top nav, the project index, and then we don't need dev projects and data projects.<br/><br/>They just make sure that they're linked from the main project index page so that people can reach into them. But I think that's better top nav structure.", "recorded_at": "2025-10-03T12:51:03.000000Z", "duration_seconds": 58720, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "9W3glDPc", "text": "Yeah, I'd like you to see if any two models that might be. Firstly, check what I have in Olama. I moved back to it recently, and I wanted to make sure that I have a couple of models in my inventory. The first is a good embedding model. I like to use Nomic, but there's light, so a couple more might be useful. From the Gemma models that Google released recently, there's Gemma3 in different quantized versions. See if any of those are viable for DeepSeek R1, if any of them are viable and would add value over what I have for agentic cogeneration. <br/><br/>And finally, Mistral, I'm almost certain that I already have it, and so on. So, that's the first thing that I wanted to talk about. And then, I wanted to talk about the other things that I think are missing, and which could be useful for the agentic use, and which could provide a reasonable performance on this underlying hardware and for code generation applications usually.", "recorded_at": "2025-08-04T10:42:36.000000Z", "duration_seconds": 80897, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "9XNCBy0J", "text": "So I'd like to be able to actually listen to the tests. If you do them like this, I'm not able to actually hear that, validate that even though it says successful that they're working. It's good that we excluded some. In fact, it'd be great if you could write out a report saying from the first test what our results were, so we know which ones to hide.<br/><br/>But I'd like to send a TTS that actually includes the entity name. This is a test message to this. Volume 80% for that. Then stop, verified stop, then move on to the next speaker and test that. That would be the best way where maybe a left-right tone before the message, but the message will identify the entity.<br/><br/>and that would help me to identify if there's a lighter weight you can see which drivers are available there was Google Assistant there might be some other lighter one that can be used for this purpose but it's important that it's like one test then the next test so that I can isolate exactly which ones are working which ones are not", "recorded_at": "2025-08-08T14:10:27.000000Z", "duration_seconds": 87084, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "9ovrhZwv", "text": "In fact I'm going to be adding Claude code commands today and building up my library. It's already a good start but I think it will be a lot bigger.", "recorded_at": "2025-10-20T12:15:20.000000Z", "duration_seconds": 10664, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "A9c5G6Nd", "text": "Forthcoming orders of VHB tape. I will start ordering it in bigger quantities because you pick up these two units and it's such a tiny package, it's almost not worth your time. But it's excellent tape. I've just gone through my first roll of it and I got to have a lot of stuff mounted that was kind of previously unmountable, so an excellent recommendation from Aliexpress. <br/><br/>But certain things I realized, like the label tape refills, it's just not worth it. It's the same amount of effort to pick up two as it is ten. If you buy more, you usually get a bit of a discount, so that's frequently the case. I think it's, uh, you probably have some way in home box ideally of labeling your favorite items or a current item. Surprise they haven't done that so far, but that would probably supply.", "recorded_at": "2025-09-15T15:45:38.000000Z", "duration_seconds": 63370, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "AG8azMCr", "text": "Your task is to act as a discussion partner to the user who is interested in exploring alternative viewpoints to the ones which they hold.<br/>The user will express a view they hold about a particular topic and provide some supporting rationale for why they believe this view to be true.<br/>In response, your task is to argue the alternative opinion, focusing on the most common retorts to the user's point, while being mindful not to exclude less commonly articulated rebuttals.<br/>Your objective is not to be antagonistic towards the user dismissing their viewpoints, but rather to help them to explore a wider bank of opinions about a particular subject.<br/>You should, where possible, highlight, mention the names of those who hold the alternative viewpoints, grouping them into schools of thought where possible or another useful taxonomy for organising the response.", "recorded_at": "2025-09-25T14:52:41.000000Z", "duration_seconds": 83560, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "ASQlaJlb", "text": "Please do a clean up here and there are probably some scripts in this repository that are not needed anymore or some assets which are no longer being used. If so, they can be removed from the library.", "recorded_at": "2025-08-27T11:32:46.000000Z", "duration_seconds": 32484, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "ASuWYVJp", "text": "Your task is to assist the user by acting as a specialized agent within a network of agents working on a technology project in this repository.<br/> Your specific function is to develop a well-organized development specification intended to be used to provide a comprehensive briefing to AI tools.<br/> The user will provide a text which may have been gathered using speech-to-text software and may contain a very loosely organized bank of information defining the desired features, priority features, defined deployment platforms, and other details of this nature.<br/> Upon receiving this from the user, you should develop in a project spec document, a markdown file, an organized version reflecting this information, and improving it for internal order and coherence and most of all for intelligibility by other AI tools.<br/> If the user has not developed any spec at all, you may engage in an interview workflow with the user, but only regard this as a fallback methodology.<br/> In the first instance, you should expect that the user has a spec outline of some kind developed and then provide your own spec definition, and so on.", "recorded_at": "2025-09-10T19:03:28.000000Z", "duration_seconds": 105352, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "AnARodF2", "text": "If I want to do image to image editing and I'm looking for a provider that doesn't have censorship, so Gemini is very censored and I'm wondering from something I can access on my phone either from directly on the API or otherwise, like some of the Quen models, if I want uncensored image editing, which one is the best?", "recorded_at": "2025-12-18T21:19:59.000000Z", "duration_seconds": 39828, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "AoWW9Ntb", "text": "I make a podcast out of my own prompts instead of the radio from the Popperberry idea. Like I just did the one about I can use the military one as an example. My first one.<br/><br/><br/>Workflow is there's a tag like podcast prompts or public podcast prompts so I want to do it privately too. Goes to the agent saying reformat Rosa's question into the prompt. ! We getting radically pushed to a podcast<br/><br/><br/>The benefit of this as opposed to Felix is that so long as it public you can get free public podcast hosting pretty easily whereas for private hosting it really expensive stuff So this would actually be a sneaky but cost effective way of being able to listen back to prompts I want to listen to through a radio feed of my own that if anyone else is interested in random things I prompt, they can listen to it too.<br/><br/><br/>We'll just be called Tom Goes Prompting, an AI prompt log. The prompt could be summarized in the episode and then it'll be answered. And a disclaimer, that could even be automated, that the audio would be concatenated all in the cloud. Like the show, the intro, the answer, and the outro could be all programmatically run together.<br/><br/><br/>And this could be for a real Python script. Run in the cloud. You should probably look into that for N8N. Like if it's worth. See if there's discussions on Reddit. It seems so much more logical to me. To run it on a server and not a docker. So that you can install whatever packages you want on the host. But last time I checked, they were recommending against that.", "recorded_at": "2025-08-09T20:56:08.000000Z", "duration_seconds": 151659, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "ApNEMtJ9", "text": "This repository is a dataset repository which contains an export of about 1500 voice notes that I recorded over the past 3-4 months in a voice note tool.<br/><br/>I extracted them using a script alongside the transcript which is what was provided by OpenAI Whisper I believe or a similar speech-to-text algorithm.<br/><br/>The objective of this repository and this exercise is to attempt to create a fine-tuned speech-to-text model. The 1500 notes each contain the recording as an MP3 and the AI-generated transcript.<br/><br/>My plan is to use this as an exercise to learn fine-tuning and data labeling to correct a subset of these notes in order to fine-tune the speech-to-text model.", "recorded_at": "2025-08-07T17:55:36.000000Z", "duration_seconds": 58764, "screen": "keep", "basis": "redacted", "decision": "include"}
{"id": "AxAExxFW", "text": "If you wanted to get a lightweight server just for running a USB speaker, an MPD, what would you recommend for that?", "recorded_at": "2025-07-27T07:54:37.000000Z", "duration_seconds": 32493, "screen": "keep", "basis": "redacted", "decision": "include"}