Ignacy Łątka

@b_nnett @ankie_me We faced the same problem in Argent and managed to take it down to <500 toks per screenshot. Take a look if you're interested: github.com/software-mansi…

@ankie_me a screenshot of the simulator uses ~2.5k tokens

iOS simulator directly in Codex!

We set out to build a simple yet flexible implementation of Radiance Cascades. 💡 ➡️𝚝𝚢𝚙𝚎𝚐𝚙𝚞/𝚛𝚊𝚍𝚒𝚊𝚗𝚌𝚎-𝚌𝚊𝚜𝚌𝚊𝚍𝚎𝚜 leverages TypeGPU composability, so you can integrate RC into your app and customize it without needing to know all the internals. ➡️ We plugged it into an existing example and got some nice GI results with minimal integration effort. Try it out yourself! ⤵️

@swmansion @NousResearch We’re also open source on GitHub: github.com/software-mansi…

@swmansion @NousResearch Argent gives your agent the capabilities to debug, profile, inspect and navigate your native iOS apps. Check out argent here: argent.swmansion.com

Argent by @swmansion now supports Hermes by @NousResearch! Learn more below 👇

@FilipKaminski Worth adding that the agent can still read all the text perfectly well despite this dramatic size reduction :)

We made Argent optimize every screenshot fed into your agent, saving 70% of tokens on each capture, with minimal data loss.

Argent agentic toolkit is out! Had fun developing, experimenting and testing - now you guys can try it out too! argent.swmansion.com

This is amazing. Argent just completed its first task in Gesture Handler - github.com/software-mansi…, which was constantly pushed back due to higher-priority issues coming in.

We just released Argent, FOSS MCP toolkit for everything from controlling to profiling and debugging your iOS apps. We've got 60 tools and 9 skills, so lots of things to test out. React Native works out of the box. Native iOS too. argent.swmansion.com

I know these pylons will break, i'll use a multi part assemly in the future. Also not proud of the bracing and the weight, but for now i just want to get this thing in the air before i optimize anything.

Part of the active sonar project, we got to have something to track after all

This was easily the most impressive hackathon we've ever attended so far, other attendees showed us their active radars, homing turrets, sonars, long range camera scanners, doppler arrays, new flight technologies, really large interceptors and probably more that i forgot about.

just finished EUDIS with @mikolajpochec we wrote firmware for a phased mic array & a solid interface to test it next step is to build the array from scratch will be working on this project more, we have a long list of things to improve and add to it

Just added 8 new SKILL.md files to @swmansion's skillset. Your agent now has everything it needs to navigate Radon IDE's MCP server like a pro. Get them here: github.com/software-mansi…

Did you know that domain vendors often fake unavailability of a domain name so that you pay them extra to "negotiate" acquiring it? It's scummy, feels like extortion. mynymbox.io doesn't do that. For more cool and (probably) legal sites check kycnot.me

Working on an app for finding the most profitable (hopefully) Lightning Network connections. For now my idea for this is to find the least optimal paths between the highest volume nodes, then fill in the gaps.

@breskin67 wow

We may not be splitting atoms yet, but we for sure have the technology for splitting expenses. Check out split.rest to stay on the bleeding edge of the latest split methods.

Also notice the marked flares aren't the only ones, this particular rocket seems to be deploying them every second or so. I marked only 2 of them because they are the least ambiguous, the most visible. And yeah, these aren't Khinzals, Iranians don't own any of them (per sources)

Just noticed Iranians started fitting some of their rockets with flares, haven't seen anyone talking about this yet Russians recently started doing this too, it resulted in a couplefold reduction of interception successrate

@BrianRoemmele I've been thinking about how this could've been implemented. My best guess is: Use high res gaussian splatting over both google streets as well as the target video. If you imagine both splattings as 3d dot graphs, you can compare their similarity. This is a bit similar to vec dbs

It is absolutely vital you know this technology exists.