Google replaced Git tags for certain source code with obtaining via Google Drive
27 by Animux | 3 comments on Hacker News.
Wednesday, August 19, 2026
Tuesday, August 18, 2026
Monday, August 17, 2026
Sunday, August 16, 2026
New top story on Hacker News: Tell HN: Cloudflare silently injects its analytics when you switch nameservers
Tell HN: Cloudflare silently injects its analytics when you switch nameservers
12 by stagas | 0 comments on Hacker News.
A few hours ago I switched my nameservers to Cloudflare in order to enable R2 bucket serving through my own subdomain, and I found out that it silently had injected a JS analytics snippet in my HTML-only JS-free site textlog.cc — I had to go to the Analytics dashboard, Add the site to the analytics and then disable the snippet. I find this approach entirely invasive, you should opt-in to features like that not have to opt-out. Just a warning out there to folks who might not be aware of this.
12 by stagas | 0 comments on Hacker News.
A few hours ago I switched my nameservers to Cloudflare in order to enable R2 bucket serving through my own subdomain, and I found out that it silently had injected a JS analytics snippet in my HTML-only JS-free site textlog.cc — I had to go to the Analytics dashboard, Add the site to the analytics and then disable the snippet. I find this approach entirely invasive, you should opt-in to features like that not have to opt-out. Just a warning out there to folks who might not be aware of this.
Saturday, August 15, 2026
Friday, August 14, 2026
New top story on Hacker News: Show HN: Mole – Deep research agent for your terminal
Show HN: Mole – Deep research agent for your terminal
11 by lajosdeme | 3 comments on Hacker News.
Doing research with agents is fun until they blow way past budget, jumble the sources, and don't even give you the best possible answer, just sound confident. And if you want to run some research task on local data - you have no idea where your data ends up after the prompt consumes it. So I built this tool: a deep-research agent with an enforced budget, verified quotes, and a privacy boundary for local data. 1. Never spend more than you budgeted (measured overshoot is 0%). 2. Every claim carries a source 3. Data stays local (give a CSV, it'll analyze it without the data ever leaving your machine) Works with most LLMs, including coding agents, subscriptions, local models, etc. It's free and open source, would appreciate all feedback!
11 by lajosdeme | 3 comments on Hacker News.
Doing research with agents is fun until they blow way past budget, jumble the sources, and don't even give you the best possible answer, just sound confident. And if you want to run some research task on local data - you have no idea where your data ends up after the prompt consumes it. So I built this tool: a deep-research agent with an enforced budget, verified quotes, and a privacy boundary for local data. 1. Never spend more than you budgeted (measured overshoot is 0%). 2. Every claim carries a source 3. Data stays local (give a CSV, it'll analyze it without the data ever leaving your machine) Works with most LLMs, including coding agents, subscriptions, local models, etc. It's free and open source, would appreciate all feedback!
Thursday, August 13, 2026
Wednesday, August 12, 2026
New top story on Hacker News: Show HN: Programmable timer web app (for gym workouts or stretching sessions)
Show HN: Programmable timer web app (for gym workouts or stretching sessions)
6 by jotaen | 1 comments on Hacker News.
Over the last couple of months, I’ve been building a timer web app for myself that I use for workout and stretching sessions. My main use-case is gym routines that consist of repeatable sequences, e.g. where you are holding certain positions for a set time (rinse and repeat). The app counts down the program, beeps, and reads the activities out loud. Two things (I suppose) are special about it: - The timers are “programmable”, so you can freely express your own routines and procedures in a declarative notation. - The app is all static (no backend): the entire program is encoded in the URL and can be bookmarked or shared/transferred via QR-code. You can check it out at https://ift.tt/sM8A5IH , optionally with a demo program pre-loaded: https://ift.tt/dtTho6l . Source code is at https://ift.tt/TKY7kwO . I’ve also written up a small behind-the-scenes on my blog: https://ift.tt/Ng60Eoz
6 by jotaen | 1 comments on Hacker News.
Over the last couple of months, I’ve been building a timer web app for myself that I use for workout and stretching sessions. My main use-case is gym routines that consist of repeatable sequences, e.g. where you are holding certain positions for a set time (rinse and repeat). The app counts down the program, beeps, and reads the activities out loud. Two things (I suppose) are special about it: - The timers are “programmable”, so you can freely express your own routines and procedures in a declarative notation. - The app is all static (no backend): the entire program is encoded in the URL and can be bookmarked or shared/transferred via QR-code. You can check it out at https://ift.tt/sM8A5IH , optionally with a demo program pre-loaded: https://ift.tt/dtTho6l . Source code is at https://ift.tt/TKY7kwO . I’ve also written up a small behind-the-scenes on my blog: https://ift.tt/Ng60Eoz
Tuesday, August 11, 2026
Monday, August 10, 2026
New top story on Hacker News: Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots
Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots
12 by HenryNdubuaku | 1 comments on Hacker News.
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges 300-700 on sub-$200 phones such as the Samsung A-Series. On the tool call and mobile device use benchmarks, Needle 2 trades wins with closest small models like LFM2.5 230M and Apple Foundation Model, at 5x to 70x smaller, both at f16 vs Needle 2 at 2bit. Needle is based on Simple Attention Networks from our paper ( https://ift.tt/Ul1ZVSA ). Edge AI has lately meant Macs and PCs, but that is just 1.5 billion of over 21 billion connected IoT devices in the world today, and in emerging markets most phones ship under $200, no NPU, cheap GPUs. These include budget phones, Raspberry Pis, microcontrollers, wearables, small robots like Reachy Mini, and connected home devices. A conventional transformer of Needle's width and depth spends 164 MFLOPs per token, and even one squeezed down to Needle's parameter count spends 87, Needle spends 70. Even on a high-end phone, an always-on assistant lives inside a power budget; every MFLOP is milliwatt-hours, and Needle spends 7x to 85x fewer of them per token than the smallest performant LLMs. More about the architecture in the link. When we structure intelligence for consumer devices as functions with typed parameters, the only hard part is mapping a messy sentence onto them; which function, with which values. Our research found that when framed that way, the problem needs no world knowledge and no open-ended prose, which is why 45M parameters suffice. Needle 2 expands to structured extraction where the schema can be passed in-place of tools and the model returns structured output. You can use Needle as a text-classification model with an enum field, as a summarization model by providing a schema that extracts key fields, everything but free-range decode. Every product has its own tool vocabulary and fine-tuning needle helps it achieve frontier-level performance on custom tasks, so using the python package ( https://ift.tt/DU8slYC ), Needle can be fine-tuned Needle on a Mac/PC in minutes to a few hours, with automated data-generation pipeline, just pass a couple samples. Nonetheless, every response carries a learned confidence score based our Cactus Hybrid technique. If above your threshold, act, below it, escalate to the cloud or bigger model. Combining Needle 2 with a private DeepSeek-v4-Flash deployment works particularly well for enterprise-level tasks at barely any cost, we can help with this setup. We have put a lot of thoughts into Needle 2 but might still be missing quite a lot, please use the playground in the provided link to test Needle and share your thoughts, always appreciated!
12 by HenryNdubuaku | 1 comments on Hacker News.
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges 300-700 on sub-$200 phones such as the Samsung A-Series. On the tool call and mobile device use benchmarks, Needle 2 trades wins with closest small models like LFM2.5 230M and Apple Foundation Model, at 5x to 70x smaller, both at f16 vs Needle 2 at 2bit. Needle is based on Simple Attention Networks from our paper ( https://ift.tt/Ul1ZVSA ). Edge AI has lately meant Macs and PCs, but that is just 1.5 billion of over 21 billion connected IoT devices in the world today, and in emerging markets most phones ship under $200, no NPU, cheap GPUs. These include budget phones, Raspberry Pis, microcontrollers, wearables, small robots like Reachy Mini, and connected home devices. A conventional transformer of Needle's width and depth spends 164 MFLOPs per token, and even one squeezed down to Needle's parameter count spends 87, Needle spends 70. Even on a high-end phone, an always-on assistant lives inside a power budget; every MFLOP is milliwatt-hours, and Needle spends 7x to 85x fewer of them per token than the smallest performant LLMs. More about the architecture in the link. When we structure intelligence for consumer devices as functions with typed parameters, the only hard part is mapping a messy sentence onto them; which function, with which values. Our research found that when framed that way, the problem needs no world knowledge and no open-ended prose, which is why 45M parameters suffice. Needle 2 expands to structured extraction where the schema can be passed in-place of tools and the model returns structured output. You can use Needle as a text-classification model with an enum field, as a summarization model by providing a schema that extracts key fields, everything but free-range decode. Every product has its own tool vocabulary and fine-tuning needle helps it achieve frontier-level performance on custom tasks, so using the python package ( https://ift.tt/DU8slYC ), Needle can be fine-tuned Needle on a Mac/PC in minutes to a few hours, with automated data-generation pipeline, just pass a couple samples. Nonetheless, every response carries a learned confidence score based our Cactus Hybrid technique. If above your threshold, act, below it, escalate to the cloud or bigger model. Combining Needle 2 with a private DeepSeek-v4-Flash deployment works particularly well for enterprise-level tasks at barely any cost, we can help with this setup. We have put a lot of thoughts into Needle 2 but might still be missing quite a lot, please use the playground in the provided link to test Needle and share your thoughts, always appreciated!
Sunday, August 9, 2026
Saturday, August 8, 2026
Friday, August 7, 2026
Thursday, August 6, 2026
Wednesday, August 5, 2026
Tuesday, August 4, 2026
Monday, August 3, 2026
Sunday, August 2, 2026
Saturday, August 1, 2026
Friday, July 31, 2026
New top story on Hacker News: Show HN: Slope remade in HTML5 to load instantly on any browser, any device
Show HN: Slope remade in HTML5 to load instantly on any browser, any device
17 by novlrdotcom | 6 comments on Hacker News.
A whole free daily game served in under 100kb with no ads, lags or logins. The URL will serve a new track every day, forever. And I've added setting sliders so players have full control over their experience. Would love to know your thoughts guys. This was just me working with Claude Opus 5. Cheers, Nathan
17 by novlrdotcom | 6 comments on Hacker News.
A whole free daily game served in under 100kb with no ads, lags or logins. The URL will serve a new track every day, forever. And I've added setting sliders so players have full control over their experience. Would love to know your thoughts guys. This was just me working with Claude Opus 5. Cheers, Nathan
Thursday, July 30, 2026
Wednesday, July 29, 2026
Tuesday, July 28, 2026
Monday, July 27, 2026
Sunday, July 26, 2026
Saturday, July 25, 2026
New top story on Hacker News: Show HN: Brolly, a plain-text weather forecast site
Show HN: Brolly, a plain-text weather forecast site
17 by jsax | 6 comments on Hacker News.
The UK MET office recently redesigned their site, adding a lot of additional whitespace, scrolling, and animations. This significantly reduced its usability for me, and left me wanting an ‘at a glance’ weather site. I made https://brolly.sh , a minimalist, plain text weather forecasting site. You can use it to view weather from around the world, with: 7 day forecast; Previous day log (so you can confirm it definitely was cooler / hotter / wetter / drier yesterday!); Hourly rain, wind, temperature, conditions; Hourly UV, air quality and pollen, including pollen type specific forecasts within the EU / UK; Location search and last 5 locations; Location specific units. I mostly made the site for myself, if anyone else also benefits from it that’s an added advantage. You can check out the weather in York, UK at https://ift.tt/eTV25yu , or search for a location at https://brolly.sh The site is deliberately styled as a single long scrollable column, to work on mobile phones. You can view it on desktop too, there's just a lot of horizontal padding. I naturally took a lot of inspiration from plaintextsports.com. Despite not being a sports fan, I love its aesthetic. But, you'll hopefully see that this site isn't a rip off, and has its own deliberate look and feel. Visualisations are really important to showing information at a glance. I spent a lot of time designing the different visualisations and making them work with only characters. My favourite is the hourly heat map used for pollen count. It's also frustrating to have an interactive site, where you can't share a page with a friend and have them see what you're seeing. To solve this, all page state (i.e. location, selected day, expanded / collapsed sections), is stored in the URL. You can share or bookmark the specific view, and know that you'll always be able to come back to it. The site uses PocketBase. It’s written in Go and plain HTML/JavaScript/CSS. All pages are backend rendered, with light JavaScript to handle re-loading content without page jumps when using interactive features like next / previous day navigation. Weather forecasts are fetched from open-meteo.com, which has a very generous free tier. However, I also built a custom LRU cache on top of PocketBase’s SQLite DB to cache forecasts for 5 minutes, and avoid putting unnecessary pressure on the open-meteo API.
17 by jsax | 6 comments on Hacker News.
The UK MET office recently redesigned their site, adding a lot of additional whitespace, scrolling, and animations. This significantly reduced its usability for me, and left me wanting an ‘at a glance’ weather site. I made https://brolly.sh , a minimalist, plain text weather forecasting site. You can use it to view weather from around the world, with: 7 day forecast; Previous day log (so you can confirm it definitely was cooler / hotter / wetter / drier yesterday!); Hourly rain, wind, temperature, conditions; Hourly UV, air quality and pollen, including pollen type specific forecasts within the EU / UK; Location search and last 5 locations; Location specific units. I mostly made the site for myself, if anyone else also benefits from it that’s an added advantage. You can check out the weather in York, UK at https://ift.tt/eTV25yu , or search for a location at https://brolly.sh The site is deliberately styled as a single long scrollable column, to work on mobile phones. You can view it on desktop too, there's just a lot of horizontal padding. I naturally took a lot of inspiration from plaintextsports.com. Despite not being a sports fan, I love its aesthetic. But, you'll hopefully see that this site isn't a rip off, and has its own deliberate look and feel. Visualisations are really important to showing information at a glance. I spent a lot of time designing the different visualisations and making them work with only characters. My favourite is the hourly heat map used for pollen count. It's also frustrating to have an interactive site, where you can't share a page with a friend and have them see what you're seeing. To solve this, all page state (i.e. location, selected day, expanded / collapsed sections), is stored in the URL. You can share or bookmark the specific view, and know that you'll always be able to come back to it. The site uses PocketBase. It’s written in Go and plain HTML/JavaScript/CSS. All pages are backend rendered, with light JavaScript to handle re-loading content without page jumps when using interactive features like next / previous day navigation. Weather forecasts are fetched from open-meteo.com, which has a very generous free tier. However, I also built a custom LRU cache on top of PocketBase’s SQLite DB to cache forecasts for 5 minutes, and avoid putting unnecessary pressure on the open-meteo API.
New top story on Hacker News: Show HN: I made some transistor animations
Show HN: I made some transistor animations
14 by stunningllama | 1 comments on Hacker News.
Hi HN, I made some animations of the most important kinds of transistors using my semiconductor simulation, details of which are on the page. I tried to make the visuals as realistic as possible while also aiming for clarity. If you want to go beyond the charge carriers and look at, for example, the electric field, you can do so in the simulation software. The desktop software also has less common devices like IBGTs and SCRs that have similar animations. The last thread about my software was posted here about a year ago: https://ift.tt/oUfy1Tk
14 by stunningllama | 1 comments on Hacker News.
Hi HN, I made some animations of the most important kinds of transistors using my semiconductor simulation, details of which are on the page. I tried to make the visuals as realistic as possible while also aiming for clarity. If you want to go beyond the charge carriers and look at, for example, the electric field, you can do so in the simulation software. The desktop software also has less common devices like IBGTs and SCRs that have similar animations. The last thread about my software was posted here about a year ago: https://ift.tt/oUfy1Tk
Friday, July 24, 2026
Thursday, July 23, 2026
Wednesday, July 22, 2026
Tuesday, July 21, 2026
Monday, July 20, 2026
Sunday, July 19, 2026
Saturday, July 18, 2026
Friday, July 17, 2026
Thursday, July 16, 2026
Wednesday, July 15, 2026
Tuesday, July 14, 2026
Monday, July 13, 2026
Sunday, July 12, 2026
Saturday, July 11, 2026
Friday, July 10, 2026
Thursday, July 9, 2026
New top story on Hacker News: Show HN: Getting GLM 5.2 running on my slow computer
Show HN: Getting GLM 5.2 running on my slow computer
45 by vforno | 3 comments on Hacker News.
A few days ago I found myself trying out GLM 5.2 and was really positively impressed. The capabilities and security I was getting from this LLM are similar to those I've gotten from models like Claude or GPT, and this really surprised me. But then I thought, "I wonder how it would work on a normal computer like mine," and above all, "I wonder if it would work without going into OOM on a computer like mine." So I started working with the help of agents to test this possibility. I started converting the model to int4, understanding MTP usage, and if possible implementing DSA for long context. How it responds in int4 and whether the quality is maintained or not. Until I got to the point, on my computer with 32GB of RAM, I was able to communicate with GLM 5.2 with times that, of course, aren't high in cold start, but even then, we're talking about 0.1 tok/s, but that wasn't important to me. The important thing was the journey to reach this goal. I just wanted it to work at all costs, even slowly. So I created Colibrì, which was born from a very simple idea, to be honest, but tested in every way, where a 744B Mixture-of-Experts model activates only ~40B parameters per token—and only ~11 GB of those change from token to token (the routed experts). So: The dense part (attention, shared experts, embeddings—~17B params) stays resident in RAM at int4 (~9.9 GB); The 21,504 routed experts (75 MoE layers × 256 experts + the MTP head, ~19 MB each at int4) live on disk (~370 GB) and are streamed on demand, with a per-layer LRU cache, an optional pinned hot-store, and the OS page cache as a free L2. The engine is a single C file (c/glm.c, ~1,300 lines) plus small headers. No BLAS, no Python at runtime, no GPU.No GPU or serious hardware because I don't have that hardware so I can't test it on hardware that is more powerful than my computer.Colibrì is a one-person project, written and tested entirely on a 12-core laptop with 25 GB of RAM — the numbers above are the ceiling of what I can measure at home. Any feedback is welcome! (and if anyone wanted to participate in the project I would be delighted) Repo: https://ift.tt/E043KxO
45 by vforno | 3 comments on Hacker News.
A few days ago I found myself trying out GLM 5.2 and was really positively impressed. The capabilities and security I was getting from this LLM are similar to those I've gotten from models like Claude or GPT, and this really surprised me. But then I thought, "I wonder how it would work on a normal computer like mine," and above all, "I wonder if it would work without going into OOM on a computer like mine." So I started working with the help of agents to test this possibility. I started converting the model to int4, understanding MTP usage, and if possible implementing DSA for long context. How it responds in int4 and whether the quality is maintained or not. Until I got to the point, on my computer with 32GB of RAM, I was able to communicate with GLM 5.2 with times that, of course, aren't high in cold start, but even then, we're talking about 0.1 tok/s, but that wasn't important to me. The important thing was the journey to reach this goal. I just wanted it to work at all costs, even slowly. So I created Colibrì, which was born from a very simple idea, to be honest, but tested in every way, where a 744B Mixture-of-Experts model activates only ~40B parameters per token—and only ~11 GB of those change from token to token (the routed experts). So: The dense part (attention, shared experts, embeddings—~17B params) stays resident in RAM at int4 (~9.9 GB); The 21,504 routed experts (75 MoE layers × 256 experts + the MTP head, ~19 MB each at int4) live on disk (~370 GB) and are streamed on demand, with a per-layer LRU cache, an optional pinned hot-store, and the OS page cache as a free L2. The engine is a single C file (c/glm.c, ~1,300 lines) plus small headers. No BLAS, no Python at runtime, no GPU.No GPU or serious hardware because I don't have that hardware so I can't test it on hardware that is more powerful than my computer.Colibrì is a one-person project, written and tested entirely on a 12-core laptop with 25 GB of RAM — the numbers above are the ceiling of what I can measure at home. Any feedback is welcome! (and if anyone wanted to participate in the project I would be delighted) Repo: https://ift.tt/E043KxO
Wednesday, July 8, 2026
Tuesday, July 7, 2026
Monday, July 6, 2026
Sunday, July 5, 2026
Saturday, July 4, 2026
Friday, July 3, 2026
New top story on Hacker News: Show HN: Bramble – Local-first password manager
Show HN: Bramble – Local-first password manager
85 by MegagramEnjoyer | 17 comments on Hacker News.
I'm currently working on Bramble, an open source password manager with P2P cross-device sync. Initially I released the Chrome extension, but recently I also published the Android app and iOS is pending Apple's approval. Besides that, the latest version also includes passkey storage for all platforms! About Bramble: It aims to be as feature-rich as all popular and a replacement for cloud-based providers. I don't think we need to store our data in the cloud and be at the whims of companies raising their prices every year. There's always a breach and then we find out that some fields aren't encrypted, metadata is visible, and so on. I'm frustrated with this and the increasing lack of transparency during these breaches. The P2P sync in Bramble uses a Nostr relay (which can be self-hosted) to keep your devices in sync. The relay just introduces the devices to each other; the data then flows directly over WebRTC, so there's no vault server and no cloud copy of your passwords anywhere. What leaves your device is end-to-end encrypted and your devices authenticate each other directly, so a snooping or MITM relay gets practically nothing. Crypto is all done in Rust so I can control exactly how key material lives and dies in memory (secrets get zeroed out, no GB leaving copies lying around). In Chromium it's a wasm module, on mobile it's native builds bridged over via uniffi. Android app: I'm still deciding whether to publish the app on Play store or simply provide the signed APK which users can sideload. Reason for that is Google's plan to lock down Android and take away ownership from its users. Read more about it here: https://ift.tt/wSP8taf The app uses no Play APIs whatsoever and runs perfectly on GrapheneOS, where I actually did all my testing. Questions, feedback, feature requests - all welcome! TL;DR: I dislike private-equity and venture funded companies messing with our security, so I created my own Password Manager which is local-first, free, open source and as transparent as it gets.
85 by MegagramEnjoyer | 17 comments on Hacker News.
I'm currently working on Bramble, an open source password manager with P2P cross-device sync. Initially I released the Chrome extension, but recently I also published the Android app and iOS is pending Apple's approval. Besides that, the latest version also includes passkey storage for all platforms! About Bramble: It aims to be as feature-rich as all popular and a replacement for cloud-based providers. I don't think we need to store our data in the cloud and be at the whims of companies raising their prices every year. There's always a breach and then we find out that some fields aren't encrypted, metadata is visible, and so on. I'm frustrated with this and the increasing lack of transparency during these breaches. The P2P sync in Bramble uses a Nostr relay (which can be self-hosted) to keep your devices in sync. The relay just introduces the devices to each other; the data then flows directly over WebRTC, so there's no vault server and no cloud copy of your passwords anywhere. What leaves your device is end-to-end encrypted and your devices authenticate each other directly, so a snooping or MITM relay gets practically nothing. Crypto is all done in Rust so I can control exactly how key material lives and dies in memory (secrets get zeroed out, no GB leaving copies lying around). In Chromium it's a wasm module, on mobile it's native builds bridged over via uniffi. Android app: I'm still deciding whether to publish the app on Play store or simply provide the signed APK which users can sideload. Reason for that is Google's plan to lock down Android and take away ownership from its users. Read more about it here: https://ift.tt/wSP8taf The app uses no Play APIs whatsoever and runs perfectly on GrapheneOS, where I actually did all my testing. Questions, feedback, feature requests - all welcome! TL;DR: I dislike private-equity and venture funded companies messing with our security, so I created my own Password Manager which is local-first, free, open source and as transparent as it gets.
Thursday, July 2, 2026
Wednesday, July 1, 2026
New top story on Hacker News: A complete ClickHouse OLAP engine, compiled to WebAssembly
A complete ClickHouse OLAP engine, compiled to WebAssembly
11 by porridgeraisin | 0 comments on Hacker News.
11 by porridgeraisin | 0 comments on Hacker News.
Tuesday, June 30, 2026
Monday, June 29, 2026
Sunday, June 28, 2026
Saturday, June 27, 2026
New top story on Hacker News: Running a software jam in a world of slop
Running a software jam in a world of slop
5 by foxmoss | 2 comments on Hacker News.
I'm Fox. I'm a 16 year old, and I've been working mostly working on making projects I thought were cool & would do well on the internet over the last year. You can check out my other blog posts if you want to get a sense of what that means: < https://foxmoss.com/blog >. Hack Club noticed these projects and thought I would be well suited to run an event. This was my reaction, I wanted to make something that could encourage the same competition as well the feedback I get from places like HN and appreciate well made projects. Hack Club does a good job at throwing money at people who make projects, but a iffy job at rewarding hard work. I wanted to change that. Radish Jam < https://radish.hackclub.com/ > was my reaction to that, and this blog post goes through my thought processes in logistics. How something similar could be run again either by another Hack Clubber or an adult looking to run something for similar for adults :)
5 by foxmoss | 2 comments on Hacker News.
I'm Fox. I'm a 16 year old, and I've been working mostly working on making projects I thought were cool & would do well on the internet over the last year. You can check out my other blog posts if you want to get a sense of what that means: < https://foxmoss.com/blog >. Hack Club noticed these projects and thought I would be well suited to run an event. This was my reaction, I wanted to make something that could encourage the same competition as well the feedback I get from places like HN and appreciate well made projects. Hack Club does a good job at throwing money at people who make projects, but a iffy job at rewarding hard work. I wanted to change that. Radish Jam < https://radish.hackclub.com/ > was my reaction to that, and this blog post goes through my thought processes in logistics. How something similar could be run again either by another Hack Clubber or an adult looking to run something for similar for adults :)
Friday, June 26, 2026
New top story on Hacker News: U.S. government will decide who gets to use GPT-5.6
U.S. government will decide who gets to use GPT-5.6
247 by alain94040 | 480 comments on Hacker News.
https://ift.tt/TpD6y5n
247 by alain94040 | 480 comments on Hacker News.
https://ift.tt/TpD6y5n
Thursday, June 25, 2026
New top story on Hacker News: Show HN: OpenKnowledge – open source AI-first alternative to Obsidian/Notion
Show HN: OpenKnowledge – open source AI-first alternative to Obsidian/Notion
35 by engomez | 9 comments on Hacker News.
Hi HN, Nick here. We’re launching OpenKnowledge ( https://ift.tt/XZUzyqa ), a “what you see is what you get” markdown editor that has direct integrations with Claude, Codex, and Cursor. Available as MacOS app or CLI. Fully free/local and OSS ( https://ift.tt/Y3kwILm ). We built this because we wanted a “Google docs” like experience for writing and sharing markdown files across our team. Obsidian is the best alternative we tried, but found it doesn’t have a true “what you see is what you get” UI and it didn’t integrate well with Claude/Codex outside of community plugins. So we built OpenKnowledge. It takes shape as: 1. A MacOS app with a file navigator, the WYSIWYG editor, and link explorer. 2. Integrations with the Claude, Codex, and Cursor desktop apps. The agents can open an OpenKnowledge editor within their embedded web browsers for a side-by-side experience. 3. Built-in mcps, skills, and RAG for LLM-wiki and “AI Second Brain” scenarios + spec writing 4. An embedded terminal and CLI for TUI-first users OSS stack includes: Tiptap/prosemirror, CodeMirror, yjs (CRDT), Electron (MacOS app), Orama, remark/rehype/micromark/mdast, @pierre/trees On the architecture side, the interesting eng. challenges included: 1. A pipeline to convert ProseMirror to markdown in a bidirectional lossless way. ProseMirror uses ASTs, which are not designed to have byte-fidelity. 2. A dual-observer CRDT to keep the ProseMirror and markdown state in-sync. The CRDT + git also power a collaborative experience that shows what Agents are doing in the markdown, have undo/redo, and version history. The “Share” and cloud-sync functionality are geared for team collaboration. They feel “no-code” but leverage git/GitHub under the hood, which also means data stays fully private. In that spirit, we made OpenKnowledge open source for anybody who’s curious or who’d like to contribute. We’re actively thinking about plugins/extensibility and what’s next. If you have suggestions or feedback, would love to hear it.
35 by engomez | 9 comments on Hacker News.
Hi HN, Nick here. We’re launching OpenKnowledge ( https://ift.tt/XZUzyqa ), a “what you see is what you get” markdown editor that has direct integrations with Claude, Codex, and Cursor. Available as MacOS app or CLI. Fully free/local and OSS ( https://ift.tt/Y3kwILm ). We built this because we wanted a “Google docs” like experience for writing and sharing markdown files across our team. Obsidian is the best alternative we tried, but found it doesn’t have a true “what you see is what you get” UI and it didn’t integrate well with Claude/Codex outside of community plugins. So we built OpenKnowledge. It takes shape as: 1. A MacOS app with a file navigator, the WYSIWYG editor, and link explorer. 2. Integrations with the Claude, Codex, and Cursor desktop apps. The agents can open an OpenKnowledge editor within their embedded web browsers for a side-by-side experience. 3. Built-in mcps, skills, and RAG for LLM-wiki and “AI Second Brain” scenarios + spec writing 4. An embedded terminal and CLI for TUI-first users OSS stack includes: Tiptap/prosemirror, CodeMirror, yjs (CRDT), Electron (MacOS app), Orama, remark/rehype/micromark/mdast, @pierre/trees On the architecture side, the interesting eng. challenges included: 1. A pipeline to convert ProseMirror to markdown in a bidirectional lossless way. ProseMirror uses ASTs, which are not designed to have byte-fidelity. 2. A dual-observer CRDT to keep the ProseMirror and markdown state in-sync. The CRDT + git also power a collaborative experience that shows what Agents are doing in the markdown, have undo/redo, and version history. The “Share” and cloud-sync functionality are geared for team collaboration. They feel “no-code” but leverage git/GitHub under the hood, which also means data stays fully private. In that spirit, we made OpenKnowledge open source for anybody who’s curious or who’d like to contribute. We’re actively thinking about plugins/extensibility and what’s next. If you have suggestions or feedback, would love to hear it.
Wednesday, June 24, 2026
New top story on Hacker News: Show HN: LookAway, a Mac break reminder that knows when not to interrupt
Show HN: LookAway, a Mac break reminder that knows when not to interrupt
5 by _kush | 0 comments on Hacker News.
Hello, I'm Kushagra and I am the indie developer behind LookAway (I've posted about it earlier but it has received quite a lot of updates since the last time so I am posting it again). LookAway is a native break reminder for macOS that doesn't interrupt. I built it because I work from home and I spend a lot of time in front of my screens. It's very easy for me to get lost in the flow and I can end up sitting for hours. Due to this, I started facing issues like eye strain and back pain by the end of the day. The solution to this was simply taking enough breaks throughout the day. But remembering to take breaks was difficult, especially when I was in the flow. I tried some reminder apps but the problem with those was that they always interrupted me at the worst moments. So I ended up not using them. LookAway is designed not to interrupt. It gives enough heads up before a break so that you're not caught off-guard. It's also context-aware and it automatically pauses when you go into a meeting, start watching a video, record screen, and much more. It even waits for you to finish typing or dictating when a break is due. One thing worth mentioning is the free iOS counterpart LookAway Mirror. When your Mac goes on a break, your iOS devices can also mirror the same break so you don't end up scrolling your phone screen during the Mac break. I've spent a lot of time in making LookAway the least annoying break reminder app and I would love to know your thoughts. It's a native Swift app so it doesn't take much resources (150MB RAM and <1% CPU when idle). It's available to download from the website (lookaway.com), Setapp, and the App Store. Thank you!
5 by _kush | 0 comments on Hacker News.
Hello, I'm Kushagra and I am the indie developer behind LookAway (I've posted about it earlier but it has received quite a lot of updates since the last time so I am posting it again). LookAway is a native break reminder for macOS that doesn't interrupt. I built it because I work from home and I spend a lot of time in front of my screens. It's very easy for me to get lost in the flow and I can end up sitting for hours. Due to this, I started facing issues like eye strain and back pain by the end of the day. The solution to this was simply taking enough breaks throughout the day. But remembering to take breaks was difficult, especially when I was in the flow. I tried some reminder apps but the problem with those was that they always interrupted me at the worst moments. So I ended up not using them. LookAway is designed not to interrupt. It gives enough heads up before a break so that you're not caught off-guard. It's also context-aware and it automatically pauses when you go into a meeting, start watching a video, record screen, and much more. It even waits for you to finish typing or dictating when a break is due. One thing worth mentioning is the free iOS counterpart LookAway Mirror. When your Mac goes on a break, your iOS devices can also mirror the same break so you don't end up scrolling your phone screen during the Mac break. I've spent a lot of time in making LookAway the least annoying break reminder app and I would love to know your thoughts. It's a native Swift app so it doesn't take much resources (150MB RAM and <1% CPU when idle). It's available to download from the website (lookaway.com), Setapp, and the App Store. Thank you!
Tuesday, June 23, 2026
Monday, June 22, 2026
Sunday, June 21, 2026
Saturday, June 20, 2026
Friday, June 19, 2026
Thursday, June 18, 2026
New top story on Hacker News: Agentic Resource Discovery Specification
Agentic Resource Discovery Specification
14 by damick | 3 comments on Hacker News.
https://ift.tt/DKX795I...
14 by damick | 3 comments on Hacker News.
https://ift.tt/DKX795I...
New top story on Hacker News: Ask HN: What is the job market like?
Ask HN: What is the job market like?
23 by gardnr | 17 comments on Hacker News.
We've all heard about layoffs; what are people's actual lived experiences when it comes to the recent changes in the job market?
23 by gardnr | 17 comments on Hacker News.
We've all heard about layoffs; what are people's actual lived experiences when it comes to the recent changes in the job market?
Wednesday, June 17, 2026
New top story on Hacker News: The hacker sent by Anthropic to calm the government's nerves about AI safety
The hacker sent by Anthropic to calm the government's nerves about AI safety
26 by Brajeshwar | 14 comments on Hacker News.
Readable: https://ift.tt/0dxbtJf...
26 by Brajeshwar | 14 comments on Hacker News.
Readable: https://ift.tt/0dxbtJf...
Tuesday, June 16, 2026
Monday, June 15, 2026
Sunday, June 14, 2026
New top story on Hacker News: Show HN: Trace – Offline Mac meeting transcripts you can flag mid-call
Show HN: Trace – Offline Mac meeting transcripts you can flag mid-call
3 by AG342 | 2 comments on Hacker News.
I'm the developer of Trace, a non-intrusive, shortcut-driven Mac app that records and transcribes your meetings on-device. I know, another meeting transcription app. Please bear with me though, I'm confident that this is at least a little novel. I primarily built Trace for myself. I'd been using MacWhisper, but there was enough fiddling before each call that I'd forget to start it and walk out of an hour-long meeting with nothing written down. So the things I cared about most were that it's quick to activate and stays out of the way. You activate Trace by pressing a global shortcut (configurable), which reveals a small bar at the bottom of your screen (there's also a keystroke and/or option to hide it entirely if you'd rather not see it at all). As I was building it I wanted to bake in a couple of workflows I'd wished for in other transcription apps. 1. Mid-meeting you can press another global shortcut to mark a "key moment" and type a note. The note shows up in the resulting transcript inline at that timestamp. I wanted to add this because I kept catching myself thinking "wait, that bit matters" in meetings and reaching to jot it down in a separate app like Obsidian, which I then needed to add context to, which took me out of the meeting. I use it all the time. If I paste the transcript into an LLM afterwards (which I find myself doing more and more these days) the important moments are flagged so it doesn't gloss over them. This is more noticeable in longer meetings with lots of topics. 2. With another keyboard shortcut you can summon a rough live recap (subtitles, basically) to quickly recap what's just been said. Trace uses standard macOS microphone and system recording APIs to capture both sides of the conversation as two separate tracks and then runs the system side through on-device diarization to identify speakers. Right now we only label them as "Speaker 1", "Speaker 2", etc but there are plans for speaker labelling in the future. You can also show a "live recap" as the call is happening to review what someone just said. All transcription models run on your machine. To be clear though, Trace doesn't do any of the summarising itself, it just produces a markdown transcript, so if you want summaries then you need to pass the output to an AI. The app is sandboxed and your audio/transcripts are never uploaded anywhere - they just exist as audio files and markdown on disk. The only network call Trace is required to make is on the first run to download the speech and speaker models (around 500MB) from Hugging Face, and after that it can be used fully offline. If enabled, a Google Calendar integration can auto-name sessions but that needs a network connection. The app is £9.99 on the macOS App Store. I've been using it every day for months now and I'm super happy with how it's improved my workflow. Feedback very welcome.
3 by AG342 | 2 comments on Hacker News.
I'm the developer of Trace, a non-intrusive, shortcut-driven Mac app that records and transcribes your meetings on-device. I know, another meeting transcription app. Please bear with me though, I'm confident that this is at least a little novel. I primarily built Trace for myself. I'd been using MacWhisper, but there was enough fiddling before each call that I'd forget to start it and walk out of an hour-long meeting with nothing written down. So the things I cared about most were that it's quick to activate and stays out of the way. You activate Trace by pressing a global shortcut (configurable), which reveals a small bar at the bottom of your screen (there's also a keystroke and/or option to hide it entirely if you'd rather not see it at all). As I was building it I wanted to bake in a couple of workflows I'd wished for in other transcription apps. 1. Mid-meeting you can press another global shortcut to mark a "key moment" and type a note. The note shows up in the resulting transcript inline at that timestamp. I wanted to add this because I kept catching myself thinking "wait, that bit matters" in meetings and reaching to jot it down in a separate app like Obsidian, which I then needed to add context to, which took me out of the meeting. I use it all the time. If I paste the transcript into an LLM afterwards (which I find myself doing more and more these days) the important moments are flagged so it doesn't gloss over them. This is more noticeable in longer meetings with lots of topics. 2. With another keyboard shortcut you can summon a rough live recap (subtitles, basically) to quickly recap what's just been said. Trace uses standard macOS microphone and system recording APIs to capture both sides of the conversation as two separate tracks and then runs the system side through on-device diarization to identify speakers. Right now we only label them as "Speaker 1", "Speaker 2", etc but there are plans for speaker labelling in the future. You can also show a "live recap" as the call is happening to review what someone just said. All transcription models run on your machine. To be clear though, Trace doesn't do any of the summarising itself, it just produces a markdown transcript, so if you want summaries then you need to pass the output to an AI. The app is sandboxed and your audio/transcripts are never uploaded anywhere - they just exist as audio files and markdown on disk. The only network call Trace is required to make is on the first run to download the speech and speaker models (around 500MB) from Hugging Face, and after that it can be used fully offline. If enabled, a Google Calendar integration can auto-name sessions but that needs a network connection. The app is £9.99 on the macOS App Store. I've been using it every day for months now and I'm super happy with how it's improved my workflow. Feedback very welcome.
Saturday, June 13, 2026
Friday, June 12, 2026
Thursday, June 11, 2026
Wednesday, June 10, 2026
Tuesday, June 9, 2026
New top story on Hacker News: Show HN: Transit-format (JSON/MessagePack) reader/writer in C
Show HN: Transit-format (JSON/MessagePack) reader/writer in C
3 by delaguardo | 0 comments on Hacker News.
Transit.c is an addition to the set of libraries to support transit data interchange format written in C11. It supports full 0.8 specification of cognitect's transit-format: JSON, JSON-Verbose and MessagePack encodings, all ground and extension types, compression via keys caching, extensibility via custom tag handlers.
3 by delaguardo | 0 comments on Hacker News.
Transit.c is an addition to the set of libraries to support transit data interchange format written in C11. It supports full 0.8 specification of cognitect's transit-format: JSON, JSON-Verbose and MessagePack encodings, all ground and extension types, compression via keys caching, extensibility via custom tag handlers.
New top story on Hacker News: Ask HN: Are you still using your Vision Pro?
Ask HN: Are you still using your Vision Pro?
12 by y1n0 | 3 comments on Hacker News.
Almost two years ago there was a thread on this (https://ift.tt/Dd7IQBk). I'm curious now that more time has passed what people think?
12 by y1n0 | 3 comments on Hacker News.
Almost two years ago there was a thread on this (https://ift.tt/Dd7IQBk). I'm curious now that more time has passed what people think?
Monday, June 8, 2026
Sunday, June 7, 2026
Saturday, June 6, 2026
Friday, June 5, 2026
New top story on Hacker News: Inside FAISS: Billion-Scale Similarity Search
Inside FAISS: Billion-Scale Similarity Search
10 by tohms | 0 comments on Hacker News.
Author here. I wrote this as a visual companion to the 2017 FAISS paper ( https://ift.tt/93xTVMl ), focused on the parts I found hardest to grok from text alone. The article covers a subset of what FAISS does, with the paper as the source of truth. NSG, FastScan, IMI are not covered here, they'll get their own articles. I'd be especially interested in feedback on: - the IVFPQ / IVFADC explanation, particularly the LUT reuse argument - whether the GPU part captures enough of the actual complexity Happy to answer questions.
10 by tohms | 0 comments on Hacker News.
Author here. I wrote this as a visual companion to the 2017 FAISS paper ( https://ift.tt/93xTVMl ), focused on the parts I found hardest to grok from text alone. The article covers a subset of what FAISS does, with the paper as the source of truth. NSG, FastScan, IMI are not covered here, they'll get their own articles. I'd be especially interested in feedback on: - the IVFPQ / IVFADC explanation, particularly the LUT reuse argument - whether the GPU part captures enough of the actual complexity Happy to answer questions.
Thursday, June 4, 2026
New top story on Hacker News: Show HN: Cost.dev (YC W21) – making agents cost-aware and cheaper to call
Show HN: Cost.dev (YC W21) – making agents cost-aware and cheaper to call
4 by akh | 0 comments on Hacker News.
We launched Infracost on HN five years ago ( https://ift.tt/EC8XeJz ) where our CLI generated cost estimates for infra-as-code, e.g. "this Terraform PR adds $400/mo". The idea was to shift cloud costs (FinOps) left, so engineers get visibility of costs before deployment and make better decisions. Earlier this year we started seeing agent traffic in our logs and it looked like coding agents were calling our CLI. But that CLI wasn't designed with coding agents in mind. We went down a philosophical rabbit hole to see if a CLI is even needed anymore given that Claude, Copilot et al. already follow best practices. Ultimately we decided to create a new CLI from the ground up with coding agents in mind for two reasons: 1. We optimized the CLI for agent callers and cut Claude's output token usage by up to 79% and API cost by up to 67% versus a bare-Claude baseline. We wrote a blog documenting our lessons on optimizing user token usage when designing a CLI, e.g. using predicate flags so the agent doesn't compose jq | python | wc pipelines, output format that strips JSON's redundant field names. The blog is here: https://ift.tt/YNReh8T... 2. With cloud costs, precision matters. Telling a coding agent "make this Terraform cost-optimized" can be expensive and lossy. You burn tokens loading code and policy context into every conversation. Your agent could make up a price and you wouldn't know because it's difficult to verify that across the ~10M price points that AWS, Azure and Google have. The CLI runs static analysis on the code, uses the latest prices from cloud vendors, and passes that context to the coding agent. So that's what we're launching today - Cost.dev: https://cost.dev/ . - It runs locally. Your code never leaves your machine, you get a fast feedback loop, and you're not burning API calls per character when you want to fetch prices. - The CLI does the deterministic work. Fetching price points, scanning the code, validating fixes. The coding agent does the natural-language part. You don't have to trust the LLM to remember the rules, and can verify it called the right CLI command. - It provides a consistent rule layer across every tool you use. Get cost estimates in your IDE and your coding agent with a single install. We support Claude Code, GitHub Copilot, Cursor, Windsurf, OpenAI Codex, Gemini CLI, as well as IDEs like VS Code and JetBrains Before we keep building more in that direction, I want to sanity-check with HN: is "agents writing IaC in prod" actually a thing yet, or am I betting on a future that's still a year out? I know software developers are using coding agents heavily, but are platform/infra folks doing that for prod too? Also, if you have any feedback on Cost.dev, I'd love to hear it!
4 by akh | 0 comments on Hacker News.
We launched Infracost on HN five years ago ( https://ift.tt/EC8XeJz ) where our CLI generated cost estimates for infra-as-code, e.g. "this Terraform PR adds $400/mo". The idea was to shift cloud costs (FinOps) left, so engineers get visibility of costs before deployment and make better decisions. Earlier this year we started seeing agent traffic in our logs and it looked like coding agents were calling our CLI. But that CLI wasn't designed with coding agents in mind. We went down a philosophical rabbit hole to see if a CLI is even needed anymore given that Claude, Copilot et al. already follow best practices. Ultimately we decided to create a new CLI from the ground up with coding agents in mind for two reasons: 1. We optimized the CLI for agent callers and cut Claude's output token usage by up to 79% and API cost by up to 67% versus a bare-Claude baseline. We wrote a blog documenting our lessons on optimizing user token usage when designing a CLI, e.g. using predicate flags so the agent doesn't compose jq | python | wc pipelines, output format that strips JSON's redundant field names. The blog is here: https://ift.tt/YNReh8T... 2. With cloud costs, precision matters. Telling a coding agent "make this Terraform cost-optimized" can be expensive and lossy. You burn tokens loading code and policy context into every conversation. Your agent could make up a price and you wouldn't know because it's difficult to verify that across the ~10M price points that AWS, Azure and Google have. The CLI runs static analysis on the code, uses the latest prices from cloud vendors, and passes that context to the coding agent. So that's what we're launching today - Cost.dev: https://cost.dev/ . - It runs locally. Your code never leaves your machine, you get a fast feedback loop, and you're not burning API calls per character when you want to fetch prices. - The CLI does the deterministic work. Fetching price points, scanning the code, validating fixes. The coding agent does the natural-language part. You don't have to trust the LLM to remember the rules, and can verify it called the right CLI command. - It provides a consistent rule layer across every tool you use. Get cost estimates in your IDE and your coding agent with a single install. We support Claude Code, GitHub Copilot, Cursor, Windsurf, OpenAI Codex, Gemini CLI, as well as IDEs like VS Code and JetBrains Before we keep building more in that direction, I want to sanity-check with HN: is "agents writing IaC in prod" actually a thing yet, or am I betting on a future that's still a year out? I know software developers are using coding agents heavily, but are platform/infra folks doing that for prod too? Also, if you have any feedback on Cost.dev, I'd love to hear it!
Wednesday, June 3, 2026
Tuesday, June 2, 2026
Monday, June 1, 2026
Sunday, May 31, 2026
Saturday, May 30, 2026
Friday, May 29, 2026
Thursday, May 28, 2026
Wednesday, May 27, 2026
Tuesday, May 26, 2026
Monday, May 25, 2026
Sunday, May 24, 2026
Saturday, May 23, 2026
New top story on Hacker News: Green card seekers must leave U.S. to apply, Trump administration says
Green card seekers must leave U.S. to apply, Trump administration says
85 by tlhunter | 370 comments on Hacker News.
https://ift.tt/mSdJua4... https://ift.tt/o4uJWh7... [pdf] https://twitter.com/DHSgov/status/2057817233200418837 , https://ift.tt/uWNBrYz https://ift.tt/tsBAMGJ https://ift.tt/WnCUMcu... , https://ift.tt/BDXCPpV
85 by tlhunter | 370 comments on Hacker News.
https://ift.tt/mSdJua4... https://ift.tt/o4uJWh7... [pdf] https://twitter.com/DHSgov/status/2057817233200418837 , https://ift.tt/uWNBrYz https://ift.tt/tsBAMGJ https://ift.tt/WnCUMcu... , https://ift.tt/BDXCPpV
Friday, May 22, 2026
Thursday, May 21, 2026
Wednesday, May 20, 2026
Tuesday, May 19, 2026
Monday, May 18, 2026
Sunday, May 17, 2026
Saturday, May 16, 2026
Friday, May 15, 2026
Thursday, May 14, 2026
Wednesday, May 13, 2026
Tuesday, May 12, 2026
New top story on Hacker News: Show HN: Statewright – Visual state machines that make AI agents reliable
Show HN: Statewright – Visual state machines that make AI agents reliable
13 by azurewraith | 4 comments on Hacker News.
Agentic problem solving in its current state is very brittle. I fell in love with it, but it creates as many problems as it solves. I'm Ben Cochran, I spent 20+ years in the trenches with full-stack Engineering, DevOps, high performance computing & ML with stints at NVIDIA, AMD and various other organizations most recently as a Distinguished Engineer. For agents to work reliably you either need massive parameter counts or massive context windows to keep the solution spaces workable. Most people are brute forcing reliability with bigger models and longer prompts. What if I made the problem smaller instead of making the model bigger? I took a different approach by using smaller models: models in the 13-20B parameter range and set them to task solving real SWE-bench problems. I constrained the tool and solution spaces using formal state machines. Each state in the machine defines which tools the model can access, how many iterations it gets and what transitions are valid. A planning state gets read-only tools. An implementation state gets edit tools (scoped to prevent mega edits) and write friendly bash tools. The testing state gets bash but only for testing commands. The model cannot physically skip steps or use the wrong tool at the wrong time. It is enforced via protocol, not via prompts. The results were more promising than I would have expected. Across multiple model families irrespective of age (qwen-coder, gpt-oss, gemma4) and the improvements were consistent above the 13B parameter inflection point. Below that, models can navigate the state machine but can't retain enough context to produce accurate edits. More on the research bit: https://ift.tt/I9GUNsm Surprisingly this yielded improvements in frontier models as well. Haiku and Sonnet start to punch above their weight and Opus solves more reliably with fewer tokens and death spirals. Fine tuning did not yield these kinds of functional improvements for me. The takeaway it seems is that context window utilization matters more than raw context size - a tightly scoped working context at each step outperforms a model given carte blanche over everything. Constraining LLMs which are non-idempotent by using deterministic code is a pattern that nobody is currently talking about. So, I built Statewright. Its core is a Rust engine that evaluates state machine definitions: states, transitions, guards and tool restrictions. Its orchestration doesn't use an LLM, just enforces the state machine. On top of that is a plugin layer that integrates with Claude Code (and soon Codex, Cursor and others) via MCP. When you activate a workflow, hooks enforce the guardrails per state automatically. The model sees 5 tools available instead of dozens, gets clear instructions for the current phase and transitions when conditions are met. Importantly it tells the model when it's attempting to do something that isn't in scope, incorrect or when it needs to try something else after getting stuck. You can use your agent via MCP to build a state machine for you to solve a problem in your current context. The visual editor at statewright.ai lets you tweak these workflows in a graph view... You can clearly see the failure paths, the retry loops and the approval gates. State machines aren't DAGs; they loop and retry, which is what agentic work actually needs. Statewright is currently live with a free tier, try it out in Claude Code by running the following: /plugin marketplace add statewright/statewright /plugin install statewright /reload-plugins Then "start the bugfix workflow" or /statewright start bugfix. You'll need to paste your API key when prompted. The latest versions of Claude may complain -- paste the API key again and say you really mean it, Claude is just being cautious here. Feedback is welcome on the workflow editor, the plugin experience, and tell me what workflows you'd want to build first. Agents are suggestions, states are laws.
13 by azurewraith | 4 comments on Hacker News.
Agentic problem solving in its current state is very brittle. I fell in love with it, but it creates as many problems as it solves. I'm Ben Cochran, I spent 20+ years in the trenches with full-stack Engineering, DevOps, high performance computing & ML with stints at NVIDIA, AMD and various other organizations most recently as a Distinguished Engineer. For agents to work reliably you either need massive parameter counts or massive context windows to keep the solution spaces workable. Most people are brute forcing reliability with bigger models and longer prompts. What if I made the problem smaller instead of making the model bigger? I took a different approach by using smaller models: models in the 13-20B parameter range and set them to task solving real SWE-bench problems. I constrained the tool and solution spaces using formal state machines. Each state in the machine defines which tools the model can access, how many iterations it gets and what transitions are valid. A planning state gets read-only tools. An implementation state gets edit tools (scoped to prevent mega edits) and write friendly bash tools. The testing state gets bash but only for testing commands. The model cannot physically skip steps or use the wrong tool at the wrong time. It is enforced via protocol, not via prompts. The results were more promising than I would have expected. Across multiple model families irrespective of age (qwen-coder, gpt-oss, gemma4) and the improvements were consistent above the 13B parameter inflection point. Below that, models can navigate the state machine but can't retain enough context to produce accurate edits. More on the research bit: https://ift.tt/I9GUNsm Surprisingly this yielded improvements in frontier models as well. Haiku and Sonnet start to punch above their weight and Opus solves more reliably with fewer tokens and death spirals. Fine tuning did not yield these kinds of functional improvements for me. The takeaway it seems is that context window utilization matters more than raw context size - a tightly scoped working context at each step outperforms a model given carte blanche over everything. Constraining LLMs which are non-idempotent by using deterministic code is a pattern that nobody is currently talking about. So, I built Statewright. Its core is a Rust engine that evaluates state machine definitions: states, transitions, guards and tool restrictions. Its orchestration doesn't use an LLM, just enforces the state machine. On top of that is a plugin layer that integrates with Claude Code (and soon Codex, Cursor and others) via MCP. When you activate a workflow, hooks enforce the guardrails per state automatically. The model sees 5 tools available instead of dozens, gets clear instructions for the current phase and transitions when conditions are met. Importantly it tells the model when it's attempting to do something that isn't in scope, incorrect or when it needs to try something else after getting stuck. You can use your agent via MCP to build a state machine for you to solve a problem in your current context. The visual editor at statewright.ai lets you tweak these workflows in a graph view... You can clearly see the failure paths, the retry loops and the approval gates. State machines aren't DAGs; they loop and retry, which is what agentic work actually needs. Statewright is currently live with a free tier, try it out in Claude Code by running the following: /plugin marketplace add statewright/statewright /plugin install statewright /reload-plugins Then "start the bugfix workflow" or /statewright start bugfix. You'll need to paste your API key when prompted. The latest versions of Claude may complain -- paste the API key again and say you really mean it, Claude is just being cautious here. Feedback is welcome on the workflow editor, the plugin experience, and tell me what workflows you'd want to build first. Agents are suggestions, states are laws.
Monday, May 11, 2026
Sunday, May 10, 2026
New top story on Hacker News: Ask HN: What Are You Working On? (May 2026)
Ask HN: What Are You Working On? (May 2026)
11 by david927 | 12 comments on Hacker News.
What are you working on? Any new ideas that you're thinking about?
11 by david927 | 12 comments on Hacker News.
What are you working on? Any new ideas that you're thinking about?
Saturday, May 9, 2026
Friday, May 8, 2026
Thursday, May 7, 2026
Wednesday, May 6, 2026
Tuesday, May 5, 2026
Monday, May 4, 2026
Sunday, May 3, 2026
New top story on Hacker News: Bad Connection: Global telecom exploitation by covert surveillance actors
Bad Connection: Global telecom exploitation by covert surveillance actors
16 by miohtama | 3 comments on Hacker News.
https://ift.tt/mUSdgi0... ( https://ift.tt/AJ2Q7bD )
16 by miohtama | 3 comments on Hacker News.
https://ift.tt/mUSdgi0... ( https://ift.tt/AJ2Q7bD )
Saturday, May 2, 2026
Friday, May 1, 2026
Thursday, April 30, 2026
Wednesday, April 29, 2026
Tuesday, April 28, 2026
Monday, April 27, 2026
Sunday, April 26, 2026
Saturday, April 25, 2026
Friday, April 24, 2026
Thursday, April 23, 2026
Wednesday, April 22, 2026
Tuesday, April 21, 2026
Monday, April 20, 2026
Sunday, April 19, 2026
Saturday, April 18, 2026
Friday, April 17, 2026
Thursday, April 16, 2026
Wednesday, April 15, 2026
Tuesday, April 14, 2026
New top story on Hacker News: Show HN: A memory database that forgets, consolidates, and detects contradiction
Show HN: A memory database that forgets, consolidates, and detects contradiction
12 by pranabsarkar | 6 comments on Hacker News.
Vector databases store memories. They don't manage them. After 10k memories, recall quality degrades because there's no consolidation, no forgetting, no conflict resolution. Your AI agent just gets noisier. YantrikDB is a cognitive memory engine — embed it, run it as a server, or connect via MCP. It thinks about what it stores: consolidation collapses duplicate memories, contradiction detection flags incompatible facts, temporal decay with configurable half-life lets unimportant memories fade like human memory does. Single Rust binary. HTTP + binary wire protocol. 2-voter + 1-witness HA cluster via Docker Compose or Kubernetes. Chaos-tested failover, runtime deadlock detection (parking_lot), per-tenant quotas, Prometheus metrics. Ran a 42-task hardening sprint last week — 1178 core tests, cargo-fuzz targets, CRDT property tests, 5 ops runbooks. Live on a 3-node Proxmox homelab cluster with multiple tenants. Alpha — primary user is me, looking for the second one.
12 by pranabsarkar | 6 comments on Hacker News.
Vector databases store memories. They don't manage them. After 10k memories, recall quality degrades because there's no consolidation, no forgetting, no conflict resolution. Your AI agent just gets noisier. YantrikDB is a cognitive memory engine — embed it, run it as a server, or connect via MCP. It thinks about what it stores: consolidation collapses duplicate memories, contradiction detection flags incompatible facts, temporal decay with configurable half-life lets unimportant memories fade like human memory does. Single Rust binary. HTTP + binary wire protocol. 2-voter + 1-witness HA cluster via Docker Compose or Kubernetes. Chaos-tested failover, runtime deadlock detection (parking_lot), per-tenant quotas, Prometheus metrics. Ran a 42-task hardening sprint last week — 1178 core tests, cargo-fuzz targets, CRDT property tests, 5 ops runbooks. Live on a 3-node Proxmox homelab cluster with multiple tenants. Alpha — primary user is me, looking for the second one.
Monday, April 13, 2026
Sunday, April 12, 2026
Saturday, April 11, 2026
Friday, April 10, 2026
Thursday, April 9, 2026
Wednesday, April 8, 2026
Tuesday, April 7, 2026
Monday, April 6, 2026
Sunday, April 5, 2026
Saturday, April 4, 2026
Friday, April 3, 2026
Thursday, April 2, 2026
Wednesday, April 1, 2026
Tuesday, March 31, 2026
New top story on Hacker News: Show HN: PhAIL – Real-robot benchmark for AI models. The gap to humans is 20x
Show HN: PhAIL – Real-robot benchmark for AI models. The gap to humans is 20x
6 by vertix | 7 comments on Hacker News.
I built this because I couldn't find honest numbers on how well VLA models actually work on commercial tasks. I come from search ranking at Google where you measure everything, and in robotics nobody seemed to know. PhAIL runs four models (OpenPI/pi0.5, GR00T, ACT, SmolVLA) on bin-to-bin order picking – one of the most common warehouse operations. Same robot (Franka FR3), same objects, hundreds of blind runs. The operator doesn't know which model is running. Best model: 64 UPH. Human teleoperating the same robot: 330. Human by hand: 1,300+. Everything is public – every run with synced video and telemetry, the fine-tuning dataset, training scripts. The leaderboard is open for submissions. Happy to answer questions about methodology, the models, or what we observed.
6 by vertix | 7 comments on Hacker News.
I built this because I couldn't find honest numbers on how well VLA models actually work on commercial tasks. I come from search ranking at Google where you measure everything, and in robotics nobody seemed to know. PhAIL runs four models (OpenPI/pi0.5, GR00T, ACT, SmolVLA) on bin-to-bin order picking – one of the most common warehouse operations. Same robot (Franka FR3), same objects, hundreds of blind runs. The operator doesn't know which model is running. Best model: 64 UPH. Human teleoperating the same robot: 330. Human by hand: 1,300+. Everything is public – every run with synced video and telemetry, the fine-tuning dataset, training scripts. The leaderboard is open for submissions. Happy to answer questions about methodology, the models, or what we observed.
Monday, March 30, 2026
Sunday, March 29, 2026
Saturday, March 28, 2026
Friday, March 27, 2026
New top story on Hacker News: Telnyx package compromised on PyPI
Telnyx package compromised on PyPI
9 by ramimac | 49 comments on Hacker News.
https://ift.tt/TKWqnr3 https://ift.tt/tpuQsVW...
9 by ramimac | 49 comments on Hacker News.
https://ift.tt/TKWqnr3 https://ift.tt/tpuQsVW...
Thursday, March 26, 2026
New top story on Hacker News: Show HN: Orloj – agent infrastructure as code (YAML and GitOps)
Show HN: Orloj – agent infrastructure as code (YAML and GitOps)
6 by An0n_Jon | 1 comments on Hacker News.
Hey HN, we're Jon and Kristiane, and we're building Orloj ( https://orloj.dev ), an open-source (Apache 2.0) orchestration runtime for multi-agent AI systems. You define agents, tools, policies, and workflows in declarative YAML manifests, and Orloj handles scheduling, execution, governance, and reliability. We built this because running AI agents in production today looks a lot like running containers before Kubernetes: ad-hoc scripts, no governance, no observability, no standard way to manage the lifecycle of an agent fleet. Everyone we talked to was writing the same messy glue code to wire agents together, and nobody had a good answer for "which agent called which tool, and was it supposed to?" Orloj treats agents the way infrastructure-as-code treats cloud resources. You write a manifest that declares an agent's model, tools, permissions, and execution limits. You compose agents into directed graphs — pipelines, hierarchies, or swarm loops. The part we're most excited about is governance. AgentPolicy, AgentRole, and ToolPermission are evaluated inline during execution, before every agent turn and tool call. Instead of prompt instructions that the model might ignore, these policies are a runtime gate. Unauthorized actions fail closed with structured errors and full audit trails. You can set token budgets per run, whitelist models, block specific tools, and scope policies to individual agent systems. For reliability, we built lease-based task ownership (so crashed workers don't leave orphan tasks), capped exponential retry with jitter, idempotent replay, and dead-letter handling. The scheduler supports cron triggers and webhook-driven task creation. The architecture is a server/worker split. orlojd hosts the API, resource store (in-memory for dev, Postgres for production), and task scheduler. orlojworker instances claim and execute tasks, route model requests through a gateway (OpenAI, Anthropic, Ollama, etc.), and run tools in configurable isolation — direct, sandboxed, container, or WASM. For local development, you can run everything in a single process with orlojd --embedded-worker --storage-backend=memory. Tool isolation was important to us. A web search tool probably doesn't need sandboxing, but a code execution tool should run in a container with no network, a read-only filesystem, and a memory cap. You configure this per tool based on risk level, and the runtime enforces it. We also added native MCP support. You register an MCP server (stdio or HTTP), Orloj auto-discovers its tools, and they become first-class resources with governance applied. So you can connect something like the GitHub MCP server and still have policy enforcement over what agents are allowed to do with it. Three starter blueprints are included (pipeline, hierarchical, swarm-loop). Docs: https://docs.orloj.dev We're also building out starter templates for operational workflows where governance really matters. First on the roadmap: 1. Incident response triage, 2. Compliance evidence collector, 3. CVE investigation pipeline, and 4. Secret rotation auditor. We have 20 templates in mind and community contributions are welcome. We're a small team and this is v0.1.0, so there's a lot still on the roadmap — hosted cloud, compliance packaging, and more. But the full runtime is open source today and we'd love feedback on what we've built so far. What would you use this for? What's missing?
6 by An0n_Jon | 1 comments on Hacker News.
Hey HN, we're Jon and Kristiane, and we're building Orloj ( https://orloj.dev ), an open-source (Apache 2.0) orchestration runtime for multi-agent AI systems. You define agents, tools, policies, and workflows in declarative YAML manifests, and Orloj handles scheduling, execution, governance, and reliability. We built this because running AI agents in production today looks a lot like running containers before Kubernetes: ad-hoc scripts, no governance, no observability, no standard way to manage the lifecycle of an agent fleet. Everyone we talked to was writing the same messy glue code to wire agents together, and nobody had a good answer for "which agent called which tool, and was it supposed to?" Orloj treats agents the way infrastructure-as-code treats cloud resources. You write a manifest that declares an agent's model, tools, permissions, and execution limits. You compose agents into directed graphs — pipelines, hierarchies, or swarm loops. The part we're most excited about is governance. AgentPolicy, AgentRole, and ToolPermission are evaluated inline during execution, before every agent turn and tool call. Instead of prompt instructions that the model might ignore, these policies are a runtime gate. Unauthorized actions fail closed with structured errors and full audit trails. You can set token budgets per run, whitelist models, block specific tools, and scope policies to individual agent systems. For reliability, we built lease-based task ownership (so crashed workers don't leave orphan tasks), capped exponential retry with jitter, idempotent replay, and dead-letter handling. The scheduler supports cron triggers and webhook-driven task creation. The architecture is a server/worker split. orlojd hosts the API, resource store (in-memory for dev, Postgres for production), and task scheduler. orlojworker instances claim and execute tasks, route model requests through a gateway (OpenAI, Anthropic, Ollama, etc.), and run tools in configurable isolation — direct, sandboxed, container, or WASM. For local development, you can run everything in a single process with orlojd --embedded-worker --storage-backend=memory. Tool isolation was important to us. A web search tool probably doesn't need sandboxing, but a code execution tool should run in a container with no network, a read-only filesystem, and a memory cap. You configure this per tool based on risk level, and the runtime enforces it. We also added native MCP support. You register an MCP server (stdio or HTTP), Orloj auto-discovers its tools, and they become first-class resources with governance applied. So you can connect something like the GitHub MCP server and still have policy enforcement over what agents are allowed to do with it. Three starter blueprints are included (pipeline, hierarchical, swarm-loop). Docs: https://docs.orloj.dev We're also building out starter templates for operational workflows where governance really matters. First on the roadmap: 1. Incident response triage, 2. Compliance evidence collector, 3. CVE investigation pipeline, and 4. Secret rotation auditor. We have 20 templates in mind and community contributions are welcome. We're a small team and this is v0.1.0, so there's a lot still on the roadmap — hosted cloud, compliance packaging, and more. But the full runtime is open source today and we'd love feedback on what we've built so far. What would you use this for? What's missing?
Wednesday, March 25, 2026
Tuesday, March 24, 2026
New top story on Hacker News: Tell HN: Litellm 1.82.7 and 1.82.8 on PyPI are compromised
Tell HN: Litellm 1.82.7 and 1.82.8 on PyPI are compromised
109 by dot_treo | 292 comments on Hacker News.
About an hour ago new versions have been deployed to PyPI. I was just setting up a new project, and things behaved weirdly. My laptop ran out of RAM, it looked like a forkbomb was running. I've investigated, and found that a base64 encoded blob has been added to proxy_server.py. It writes and decodes another file which it then runs. I'm in the process of reporting this upstream, but wanted to give everyone here a headsup. It is also reported in this issue: https://ift.tt/aU8hBTd
109 by dot_treo | 292 comments on Hacker News.
About an hour ago new versions have been deployed to PyPI. I was just setting up a new project, and things behaved weirdly. My laptop ran out of RAM, it looked like a forkbomb was running. I've investigated, and found that a base64 encoded blob has been added to proxy_server.py. It writes and decodes another file which it then runs. I'm in the process of reporting this upstream, but wanted to give everyone here a headsup. It is also reported in this issue: https://ift.tt/aU8hBTd
Monday, March 23, 2026
Sunday, March 22, 2026
Saturday, March 21, 2026
New top story on Hacker News: Show HN: Termcraft – terminal-first 2D sandbox survival in Rust
Show HN: Termcraft – terminal-first 2D sandbox survival in Rust
4 by sebosch | 0 comments on Hacker News.
I’ve been building termcraft, a terminal-first 2D sandbox survival game in Rust. The idea is to take the classic early survival progression and adapt it to a side-on terminal format instead of a tile or pixel-art engine. Current build includes: - procedural Overworld, Nether, and End generation - mining, placement, crafting, furnaces, brewing, and boats - hostile and passive mobs - villages, dungeons, strongholds, Nether fortresses, and dragon progression This is still early alpha, but it’s already playable. Project: https://ift.tt/tC5QqoM Docs: https://pagel-s.github.io/termcraft/ Demo: https://youtu.be/kR986Xqzj7E
4 by sebosch | 0 comments on Hacker News.
I’ve been building termcraft, a terminal-first 2D sandbox survival game in Rust. The idea is to take the classic early survival progression and adapt it to a side-on terminal format instead of a tile or pixel-art engine. Current build includes: - procedural Overworld, Nether, and End generation - mining, placement, crafting, furnaces, brewing, and boats - hostile and passive mobs - villages, dungeons, strongholds, Nether fortresses, and dragon progression This is still early alpha, but it’s already playable. Project: https://ift.tt/tC5QqoM Docs: https://pagel-s.github.io/termcraft/ Demo: https://youtu.be/kR986Xqzj7E
New top story on Hacker News: Show HN: Joonote – A note-taking app on your lock screen and notification panel
Show HN: Joonote – A note-taking app on your lock screen and notification panel
12 by kilgarenone | 3 comments on Hacker News.
I finally built this app after many years of being sick of unlocking my phone every goddamn time I need to take or view my notes. It particularly sucks when I'm doing my grocery and going down the list. I started building last year June. This is a native app written in Kotlin. And since I'm a 100% Web dev guy, I gotta say this wouldn't have been possible without this AI to assist me. So this isn't "vibe-coded". I simply used the chat interface in Gemini website, manually copy paste codes to build and integrate every single thing in the app! I used gemini to build it just because I was piggybacking on my last company's enterprise subscription. I personally didn't subscribe to any AI (and still don't cuz the free quota seems enough for me :) So I certainly have learnt alot about Android development, architecture patterns, Kotlin syntax, and obeying Google's whims. Can't say I love it all, but for the sake of this app, I will :) Anyway, I finally have the app I wish existed, and I'm using it everyday. It not only does the main thing I needed it to do, but there's also all this stuff: - Make your notes private if you don't want to show them on lock screen. - Create check/to-do lists. - Set one time or recurring reminders. - Full-text search your notes in the app. - Speech-to-text. - Organize your notes with custom or color labels. - Pin the app as a widget on your home screen. - You can auto backup and restore your notes on new install or Android device. - Works offline. - And no funny business happening in the background https://ift.tt/52LKWbM It's 30-day trial, then a one-time $9.99 to go Pro forever. I would love you all to check it out, FWIW. Ok thanks!
12 by kilgarenone | 3 comments on Hacker News.
I finally built this app after many years of being sick of unlocking my phone every goddamn time I need to take or view my notes. It particularly sucks when I'm doing my grocery and going down the list. I started building last year June. This is a native app written in Kotlin. And since I'm a 100% Web dev guy, I gotta say this wouldn't have been possible without this AI to assist me. So this isn't "vibe-coded". I simply used the chat interface in Gemini website, manually copy paste codes to build and integrate every single thing in the app! I used gemini to build it just because I was piggybacking on my last company's enterprise subscription. I personally didn't subscribe to any AI (and still don't cuz the free quota seems enough for me :) So I certainly have learnt alot about Android development, architecture patterns, Kotlin syntax, and obeying Google's whims. Can't say I love it all, but for the sake of this app, I will :) Anyway, I finally have the app I wish existed, and I'm using it everyday. It not only does the main thing I needed it to do, but there's also all this stuff: - Make your notes private if you don't want to show them on lock screen. - Create check/to-do lists. - Set one time or recurring reminders. - Full-text search your notes in the app. - Speech-to-text. - Organize your notes with custom or color labels. - Pin the app as a widget on your home screen. - You can auto backup and restore your notes on new install or Android device. - Works offline. - And no funny business happening in the background https://ift.tt/52LKWbM It's 30-day trial, then a one-time $9.99 to go Pro forever. I would love you all to check it out, FWIW. Ok thanks!
Friday, March 20, 2026
New top story on Hacker News: Show HN: I made an email app inspired by Arc browser
Show HN: I made an email app inspired by Arc browser
4 by johndamaia | 2 comments on Hacker News.
Email is one of those tools we check daily but its underlying experience didn’t evolve much. I use Gmail, as probably most of you reading this. The Arc browser brought joy and taste to browsing the web. Cursor created a new UX with agents ready to work for you in a handy right panel. I use these three tools every day. Since Arc was acquired by Atlassian, I’ve been wondering: what if I built a new interface that applied Arc’s UX to email rather than browser tabs, while making AI agents easily available to help manage emails, events, and files? I built a frontend PoC to showcase the idea. Try it: https://demo.define.app I’m not sure about it though... Is it worth continuing to explore this idea?
4 by johndamaia | 2 comments on Hacker News.
Email is one of those tools we check daily but its underlying experience didn’t evolve much. I use Gmail, as probably most of you reading this. The Arc browser brought joy and taste to browsing the web. Cursor created a new UX with agents ready to work for you in a handy right panel. I use these three tools every day. Since Arc was acquired by Atlassian, I’ve been wondering: what if I built a new interface that applied Arc’s UX to email rather than browser tabs, while making AI agents easily available to help manage emails, events, and files? I built a frontend PoC to showcase the idea. Try it: https://demo.define.app I’m not sure about it though... Is it worth continuing to explore this idea?
Thursday, March 19, 2026
New top story on Hacker News: Show HN: Dumped Wix for an AI Edge agent so I never have to hire junior staff
Show HN: Dumped Wix for an AI Edge agent so I never have to hire junior staff
8 by axotopia | 10 comments on Hacker News.
I run a building design consultancy. I got tired of paying Wix $40/month for a brochure that couldn’t answer simple service questions, and me wasting hours on the same FAQs. So I killed it all and spent 4 months building a 'talker': https://axoworks.com The stack is completely duct-taped: Netlify’s 10s serverless timeout forced me to split the agent into three pieces: Brain (Edge), Hands (Browser), and Voice (Edge). I haven’t coded in 30 years. This was 3 steps forward, 2 steps back, heavily guided by AI. The fight that proved it worked: 2 weeks ago, a licensed architect attacked the bot, trying to prove my business model harms the profession. The AI (DeepSeek-R3) completely dismantled his arguments. It was hilariously caustic. Log: https://ift.tt/CWQqdo6... A few battle scars: * Web Speech API works fine, right up until someone speaks Chinese without toggling the language mode. Then it forcefully spits out English phonetic gibberish. Still a headache. * Liability is the killer. Hallucinate a building code clause? We’re dead. Insurance won’t touch us. * We publish the audit logs to keep ourselves honest and make sure the system stays hardened. Audit: https://ift.tt/aEBm0k3 The hardest part was getting the intent right: making one LLM pivot seamlessly from a warm principal’s tone with a homeowner, to a defensive bulldog when attacked by a peer. That took 2.5 months of tuning. We burn through tokens with an 'Eager RAG' hack (pre-fetching guesses) just to improve responsiveness. I also ripped out the “essential” persistent DBs—less than 5% of visitors ever return, so why bother? If a client drops mid-query, their session vanishes. No server-side queues. The point: To let me operate with a network of seasoned pros, and trim the fat. Try to break it. I’ll be in the comments. Kee
8 by axotopia | 10 comments on Hacker News.
I run a building design consultancy. I got tired of paying Wix $40/month for a brochure that couldn’t answer simple service questions, and me wasting hours on the same FAQs. So I killed it all and spent 4 months building a 'talker': https://axoworks.com The stack is completely duct-taped: Netlify’s 10s serverless timeout forced me to split the agent into three pieces: Brain (Edge), Hands (Browser), and Voice (Edge). I haven’t coded in 30 years. This was 3 steps forward, 2 steps back, heavily guided by AI. The fight that proved it worked: 2 weeks ago, a licensed architect attacked the bot, trying to prove my business model harms the profession. The AI (DeepSeek-R3) completely dismantled his arguments. It was hilariously caustic. Log: https://ift.tt/CWQqdo6... A few battle scars: * Web Speech API works fine, right up until someone speaks Chinese without toggling the language mode. Then it forcefully spits out English phonetic gibberish. Still a headache. * Liability is the killer. Hallucinate a building code clause? We’re dead. Insurance won’t touch us. * We publish the audit logs to keep ourselves honest and make sure the system stays hardened. Audit: https://ift.tt/aEBm0k3 The hardest part was getting the intent right: making one LLM pivot seamlessly from a warm principal’s tone with a homeowner, to a defensive bulldog when attacked by a peer. That took 2.5 months of tuning. We burn through tokens with an 'Eager RAG' hack (pre-fetching guesses) just to improve responsiveness. I also ripped out the “essential” persistent DBs—less than 5% of visitors ever return, so why bother? If a client drops mid-query, their session vanishes. No server-side queues. The point: To let me operate with a network of seasoned pros, and trim the fat. Try to break it. I’ll be in the comments. Kee
Subscribe to:
Posts (Atom)