说实话我觉得现在扎堆卷个人助理 Agent 没什么意义... 首先用户为什么不用 Muse 呢，其次 Google 和 Apple 随时可以不需要用户授权默认拿到最高权限亲自下场来做这个事，所有的个人助理 Agent 都得死....

说来惭愧，我到现在也没用过 OpenClaw 龙虾类的产品。OpenClaw 刚一出来的时候，我就感觉像是一波流，但那时候太热了，好像没人会质疑。再说，侵入感太强了，我也是真不敢放自己设备上用。 当然，现在也可以说我是马后炮。 但大厂去做，并不是没有价值，至少把团队都打醒了，激活了战斗力，做产品的感觉也找到了。从这个角度上说，没有失败。

deepseek harness 桌面端发布了，我已经打入了他们飞书群。 dshdesktop.com/en/

Şu olay aynı Kubernetes devrimine benziyor ama pek tınlayan olmadı. Potansiyeli inanılmaz. github.com/google/ax

今天这种对话列表式的智能体UI，不出一年就会被新的形态代替，我不知道新的形态会是什么，但是今天这种形态已经开始明显的制约了我的效率，而我同时活跃的会话也就是10个这个量级而已

🚀 LLGo v1.0.4 just shipped — and it's a big one for WebAssembly. For the uninitiated: LLGo is a Go compiler built on LLVM 22. Your Go 1.20+ code compiles as-is, and you can call C/C++/Python libraries directly through the C ABI — think PyTorch, pandas, numpy — without cgo overhead. Highlights of v1.0.4: 🧩 WebAssembly milestone: the W1→W3 series defines supported wasm profiles, completes host integration + reflection, lands a single-worker runtime, and passes full compatibility acceptance. Go on wasm, for real. 🛡️ Runtime hardening: fixed reentrant weak-cleanup deadlocks and BDWGC callback reentrancy; recovered panics now keep their source locations. ⚡ Faster compiles & codegen: unsigned mul-overflow lowered as an intrinsic, constant map[K][]T vars lowered to static slices, plus an SSA alloca-builder cache. 🎯 Go language compat: generic range-over-func defers preserved; x/tools v0.50.0 brings generic receiver closures. 🧰 DX: new llgo env command, better LLGoFiles docs, native ARM64 ESP tools on Windows. github.com/xgo-dev/llgo/rele… #golang #LLVM #WebAssembly #opensource

Today is officially available in China. Your personal AI can prepare what you need before a meeting, help organize a trip before you leave, and stay on top of longer-running matters across learning, health, and personal finances. Download Today: todayai.cn/downloads

有大模型中转站，怎么没有 VPN 中转站？

Maka 的 Maintainer 群是我见过的最高质量的 AGENT 知识群，每天大家都会讨论很多设计和知识。 另外很开心，Maka 有几个 Maintainer 靠着 Maka 的知识和实践拿到了 AGENT 方向的offer。 Community over code. 很开心有这样一个社群。

dress for the job you want

We’re live. Meet TenPayGo 👋 Coming to China for travel? TenPayGo is a payment app built by Weixin Pay for overseas visitors to China. Pay wherever Weixin Pay is accepted, using Apple Pay, international cards and supported e-wallets. Sign up with your email and you’re ready to go. And there’s more — from translation to metro and bus travel, TenPayGo is built to make everyday life in China easier. Easy Pay with TenPayGo. Explore China with TenPayGo.🇨🇳

good one

MiMo-V3 is getting a new architecture. The core of it, HySparse2, is out today. Less prefill, a smaller KV cache, better long-context retrieval—and we got all three at once. Compared with MiMo-V2.6's Hybrid SWA architecture: • 5.02× lower prefill FLOPs at 1M tokens • 4.5× smaller KV cache at 1M tokens • Better MRCRv2 and RULER-v2 scores, plus lower AgentPPL and LongPPL Why build a new architecture? Agentic inference is a very different workload. Each round, a short action can return a long observation that needs to be prefilled, while the context keeps growing. That puts prefill cost, KV-cache size, and retrieval accuracy on the critical path at the same time. HySparse2 tackles all three with two levels of KV sharing: • KV Bridging: Following YOCO, full-attention layers in the cross-decoder build their K/V from self-decoder hidden states. • KV Reuse: Within each hybrid block, sparse layers reuse the preceding full-attention layer's KV cache and selection indices. Two more changes: token-level selection replaces block-level selection, and a forced window of recent tokens replaces the separate SWA branch, so local and global tokens share one KV cache. Since all cross-decoder KV caches now come from the self-decoder, prefill can stop once the self-decoder finishes. Paper: arxiv.org/pdf/2609.26368