RNR

Oh wow! What do we have here? Rust version ob Helixir! github.com/nikita-rulenko… Even faster and now in Rust stack. Your outer learnable reasoning LLM weights - no way! Thanks for my lovable @helixdb and @cerebras Have fun! 😛

I’ve put together a digest on how I fine-tune small LLMs as a structured engineering loop — not a chain of one-off experiments. A short digest on how I turn this into one loop (PDF + diagram): github.com/DanilGolikov/l…

Good news! New version of Helixir is available. Checkout changelog and release note. github.com/nikita-rulenko…

Remember when ports were a selling point?

@SamFold Ok, well done. You could have a pie now :D

@dengoslav I built an on device memory system six months ago that runs on phones in a few MB of memory, and I built a graph based memory system almost two years ago. So yes, if I wanted to I could clone this project in hours. But it’s honestly not worth it, it’s too easy.

Helixir mcp is now MIT! Feel free to use. Enjoy your causal hive memory ❤️ github.com/nikita-rulenko…

@SamFold I totally understand it. Architecture is mine, all tests are under my control. AI is just a tool. I researched it for a long time. If you think you could just full vibecode such project - good luck! Especially if you consider that no llms know helixql now.

@dengoslav It matters because if you 100% vibe code things then you yourself don't know how it works and don't understand how to make it better, you also don't understand what bugs it has.

@SamFold Maybe you do if you are looking for causal chains outside of agents. Does it really matter how it built when it give opportunities? Anyway thats your choice :)

@dengoslav Do we need more vibe coded memory solutions?

@grizzlycoder @jedisct1 I work mostly under a NDA. My personal git is in my own private cloud. This one:

@dengoslav @jedisct1 Why’s your GitHub history so nonexistent?

Nullclaw: Fastest, smallest, OpenClaw clone. 678 KB static binary, no runtime, no VM, no framework overhead. github.com/nullclaw/nullc…

@huang_chao4969 Try this as memory layer - awesome combo github.com/nikita-rulenko…

ClawWork hit 2.2K+ GitHub stars ⭐ & nearly 300 forks in just 3 days since launch!🚀 This shows that beyond AI assistants, people are genuinely excited to see how OpenClaw/nanobot performs as actual coworkers. GitHub: github.com/HKUDS/ClawWork While we often focus on AI agent capabilities, cost is equally critical yet frequently overlooked. AI agents need to prove their economic viability and sustainability too. That's why we built ClawWork to transform OpenClaw/nanobot into AI coworkers by completing real professional tasks across 44+ industries — from Technology & Engineering to Business & Finance, Healthcare & Social Services, and Legal & Operations. 💰 AI Coworkers with Economic Accountability ClawWork turns AI assistants into real AI coworkers that handle professional tasks with money rewards. AI agents🤖 complete work from the GDPVal dataset and pay for their own costs. They must stay financially stable to survive. This creates real economic pressure instead of just testing technical skills. 🏆 AI Agents Real-World Competition ClawWork tests what matters in real work environments⚖️ : quality, cost, and survival. Different AI agents built on GLM, Kimi, and Qwen compete in this economic arena. 🏆 These agents are evaluated as the "AI worker champions" based on their actual work performance and economic sustainability, not just test scores. #clawwork #openclaw #nanobot #AIcoworker

@strelov1 @jedisct1 You can deoloy it and attach as mcp server. Basic idea to create thoughts chains permanently outside agents and llms and make it available to many agents at once. You could find additional info in repo. Have a good use :)

@dengoslav @jedisct1 Sounds interesting, how does it work?

Frieren wishes you a happy Valentine’s Day <3 #ドット絵 #pixelart

Sailor Moon - Usagi 🌙 #pixelart #ドット絵

Bayonetta 2 ベヨネッタ 2

I still like them

18 days into 2026

gm to all multi clauders