Looking forward to check this out

Cooking dinner while I keep an eye on RoboClaw's PRs. @roboclawbot is the @openclaw Foundation's agent. This week a few Clawtributors like myself share the multiplayer server with the Foundation team for the first time. It feels good, with some minor hickups and I'm having a blast. Morale is high. By Friday we hope Control UI and the guest experience are a lot better.

🚀 Introducing DeepInfer, our in-house inference framework, now fully open source! Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable. 🔗 github.com/deepseek-ai/DeepI… 1/4

🦞 OpenClaw v2026.10.5 is out! 🤖 Sidebar keeps your agents within reach ↩️ Skill Workshop 2.0 gets Undo 🔗 Public conversation links need no sign-in 🎉 2,250 PRs and 637 contributors. Just the highlights. Much more in the full notes 👇 docs.openclaw.ai/releases/20…

The memory unit economics of Apple Silicon are insane. When you realise LLM inference is massively memory-bound (even at high concurrency!), you find that $ per memory bandwidth ($/GB/s) is the most important metric for inference. M5 Ultra: $5499, 96GB @ 1.2TB/s, $4.5/GB/s DGX Spark: $6950, 128GB @ 273GB/s, $25.5/GB/s DGX Station: ~$150K, 252GB @ 7.1TB/s (HBM3e), $21.1/GB/s M5 Ultra has 5.6x better $/GB/s than DGX Spark and 4.7x better $/GB/s than DGX Station. I am surprised Apple didn't increase their prices more (maybe they will so might be a good time to buy...) I expect 512GB M5 Ultra Mac Studio to start at $15899 (following the usual +$25/GB incremental memory price). That's a crazy good price for that much high speed memory. With RDMA over Thunderbolt 5 you get single digit microsecond latency and you can connect up to 5 Thunderbolt 5 cables between Macs for 400Gb/s network bandwidth (same as ConnectX-7). Thunderbolt 5 cables are ~$25 each. The main problem is the software on Apple Silicon is not great. There are about 30 competing inference engines, each supporting a small set of models, and quite unstable. The ecosystem around MLX is immature and limited (e.g. no Dynamo, LMCache, llm-d, ...). In our experience, more serious customers like Governments and Enterprises are put off by that and want something stable with an ecosystem of tools and software to run inference in production.

Sunday on the menu: my daughter with a precision screwdriver, putting her computer back together, one screw at a time. Not a single agent involved in this build ahahah The next generation of builders starts at the bench ;)

🔥OMARCHY M KERNEL RELEASE NOTES Money for Nothing And your chips for free! We are about to launch and the team is HYPED Too many people to thank in one X post. Having the time of my life working with this crew!

Tibo resetting limits today Anything left on my limits before then is just wasted tokens. Reset lands by EOD PST, so whatever's left gets burned tonight. Fast mode, Ultra, no regrets 😂 Have a goal for 16 days on one run and still going 😂

I just want to own a cybertruck in Portugal, this can be US only ahahha

Adapted @poteto's pstack and most of Cursor's plugin skills to run in Codex, or in anything that happens to run on a Codex runtime 😉 @openclaw Each skill is mirrored from a pinned upstream snapshot with a small overlay, so it behaves natively outside Cursor and doesn't drift. Install just the Cursor pack: npx skills add github.com/TheAngryPit/TheAn… @poteto, would love for you to take a look at my interpretation 🙂

Asking for the weather and getting an expired TLS certificate instead is a very literal kind of forecast 😂 Two more of my fixes made it into Hermes Agent, both cherry-picked from my PRs. The TUI /weather widget depended on wttr.in, whose certificate had expired, so every lookup failed. It now uses Open-Meteo's credential-free APIs, with TLS verification kept intact. And in Desktop Bot Mode, /resume from a Bot tab could open the session in the main Sessions view or leave the empty Bot draft on screen. The picker now keeps the Bot workspace it was opened from. PRs #111818, #96364. Thanks @Teknium and @NousResearch for bringing both in with my authorship intact 🙂

Reformatting a config file without changing a single setting was apparently enough for OCM to decide it belonged to someone else 😂 My fix shipped in OCM v0.2.48. OCM now reads LaunchAgent ownership as an actual plist value, so formatting-only edits and binary plists stay manageable. Foreign, missing or malformed owners are still refused, and systemd behaviour is unchanged. PR #149. Thanks Peter and @openclaw for getting it merged 🙂

Brain context was loading 0 bytes in every skill because it called an MCP tool name as a CLI command. Very quiet failure 😂 My fix shipped in gstack v1.91.13.0. gstack now uses the working gbrain list read and reports failed queries, and saves get verified instead of assumed. PR #3000 (landed in #3013). Thanks @garrytan for folding it into the bug-wave release and keeping me as co-author 🙂

A brain with a perfectly good local extraction model, being told it had no model, because the check only looked at env vars. Bold strategy 😂 My fix made it into GBrain v0.60.20.0. Harvest now uses the extraction model the brain actually resolves, so a database-configured local model on a keyless install isn't skipped as "keyless" anymore. My PR was a draft that conflicted with the new base. Garry reimplemented it from my src/test changes and extended it to OpenClaw compaction and the facts backstop, with a Postgres E2E. I'm credited as co-author. PR #5736 (landed in #5792). Thanks @garrytan for carrying it the rest of the way 🙂

Today I am not very happy codexing @OpenAIDevs codex is painfully slow today I am having in my pro accounts 15tokens/s no matter the os and the harness @jxnlco @thsottiaux I know you have top-notch client support and always care for us, so I hope this will be fix asap

Depending on your plans, using the one with the larger spend makes a huge difference in the unified options.

Holy cow

Introducing Muse Wonder, our new family of frontier models with state of the art performance for personal and professional agentic work.

My Bot Mr Nightly edits a daily newspaper custom made for me with all my personal/professional/family endeavors. I wake up and it's already printed to read while I drink my morning expresso