On October 21, the Deputy Head of the State Customs Service of Kyrgyzstan announced that, under the revised Eurasian Economic Union Customs Code, containers will henceforth be managed as transport equipment rather than goods; consequently, within Kyrgyzstan, they require no customs declaration, are exempt from customs fees, have no stay limits, and only need to provide a container number. Separate declarations are required only for buy-sell transactions. Previously, the Eurasian Economic Union had already allowed international freight containers to be freely used for domestic transport, transshipment, multimodal transport, and export.
On June 3, 2026, the White House signed Executive Order No. 14411, "Strengthening Customs Enforcement," shifting import regulation from cargo control to full lifecycle qualification management of importing entities.
Starting September 18, 2026, US Customs CBP will initiate IOR (Importer of Record) authenticity verification based on Executive Order No. 14411, strictly checking 6 items of information on Form 5106 including physical address, email, and phone number; virtual shell IORs will be invalidated, leading to inability to clear customs, port congestion, or cargo detention; the new regulation forces freight forwarders to use US entities with real operational capabilities and pay deposits, costs which will be passed on to sellers; foreign entity IORs must operate through CTPAT-certified customs brokers; Amazon US sellers need to immediately confirm the nature of the IOR entity and the qualifications of the customs broker with their freight forwarders.
German e-commerce platform Otto has launched a plugin on ChatGPT, allowing users to query over 19,000,000 products on the site (including home appliances and coffee machines) within the chat interface. While inquiries are handled within ChatGPT, placing orders and installation services still require returning to Otto's own platform.
This article introduces how to access amazon.com/clp/asin (replacing ASIN with the target product number) to observe Amazon's multi-dimensional recommendation groupings for a single product, thereby identifying consumers' true purchase reasons, such as demand nodes like 2-step design, safety handrail, and high weight capacity. It emphasizes that CLP does not reflect official intent labels but rather observable product relationship clues, which can be used to discover unmet cross-demands. It also suggests combining AI tools (such as SellerSprite MCP) to cluster expanded keywords into demand groups, transforming vague product directions (such as home-oriented Step Ladders) into verifiable hypotheses regarding specific scenarios, materials, colors, and usage spaces.
AMD and Intel announced a collaboration on PyTorch optimizations for x86 CPUs, integrating their respective optimizations (including AMD's ZenDNN) into the open-source library oneDNN, making it the default open-source library for x86 acceleration within PyTorch; both parties are also jointly promoting the AI Compute Extensions (ACE) instruction set for x86, focusing on accelerating matrix multiplication, with plans to integrate it into PyTorch via oneDNN.
lyc575757 released MFlare, an independently developed macOS menu bar tool. It supports aggregating usage data for Workers, R2, D1, KV, and Durable Objects via Cloudflare OAuth login, and pushes reminders when reaching 80% or 100%. It also monitors Worker deployment start/success/failure and provides access to build logs. The Free version is limited to a single account, while the Pro version ($3 one-time purchase) supports multiple accounts and MCP integration.
Vitals is a system monitoring application for macOS that offers three viewing modes: standalone window, menu bar resident, and MCP service. It categorizes resource consumption—including CPU, memory, disk, network, GPU, battery, and temperature—by app, supporting historical curves, anomaly alerts, and CSV export. With MCP protocol integration, it allows AI assistants such as Claude and Cursor to query system status via natural language. The app uses a one-time purchase model, with an initial launch price of $9.
Droid-Containers is a user-space container tool that runs Docker images without root on old Android phones. It uses proot to implement rootfs switching and zstd unpacking, supporting images such as alpine/debian/ubuntu/python/node/postgres/mysql/redis/nginx/mosquitto. It includes built-in templates and offers APK downloads; it requires Android 8+, arm64-v8a or x86_64 architecture, does not support armeabi-v7a, and is licensed under PolyForm Noncommercial 1.0.0.
aitobox has open-sourced the ATBGamePort skill (https://github.com/atbfun/ATBGamePort), which supports automatically porting EXE or Flash mini games to browser-runnable HTML5 games via Agent calls; tests show it works on Linux and macOS, with Claude Sonnet 5.5 achieving the highest fidelity, consuming about 5% of a $20 API package per game; demos for Zuma, Cut the Rope, and Kingdom Rush have been released.
Koi is a Mac app that integrates speech-to-text (Fn+⌥, unlimited offline use), a screen-aware AI assistant (Fn+⌘, capable of recognizing and interacting with on-screen content), and call recording and summarization without bot intervention (Fn+⌃); all outputs are saved locally on the Mac as pages; a 3-day free trial is offered.
According to Reuters, three major Chinese internet companies have collectively ordered over 3 million NVIDIA RTX PRO 5500 workstation graphics cards, with a potential value exceeding $35 billion; this card is based on the GB202 chip, features 84GB of memory and 1398GB/s bandwidth, supports MIG partitioning, and is optimized specifically for AI inference; due to U.S. restrictions on exports of data center chips such as B300 and H200, the RTX PRO 5500 has become a compliant high-performance alternative; NVIDIA stated that global production in 2027 may not exceed 3 million units, with actual deliveries to China possibly accounting for only 5%–10%; the Ministry of Industry and Information Technology has proposed approving procurement plans by some enterprises.
Runner 0.17.0 has been released, adding support for multi-account switching in Claude Code. Multiple accounts (e.g., Personal, Work) can be added under Settings → Agents → Claude Code Accounts, each retaining its own login status, usage data, and local configuration; after switching accounts, original conversations can be continued by launching via CLAUDE_CONFIG_DIR to reuse skills, MCP, and plugins from the main account; supports macOS (Apple Silicon) and Windows x64 (10 1809+); the project is independently developed by author yicheng47, with the open-source address at https://github.com/yicheng47/runner.
Agent Pets is a browser extension that turns Claude Code, OpenCode, and omp sessions into agent pets walking along the bottom of the screen, supporting cross-tab command approvals and replies; it runs locally, requires no account or cloud services, and costs $6 for a one-time purchase.
Inca is a free, open-source 3D application supporting polygon modeling, rigging, animation, shading, rendering, and script extensions. Its interface and hotkeys fully replicate Maya 2020. It provides Windows (portable exe), Linux (AppImage), and macOS (dmg) versions with no installation dependencies. Licensed under MIT, it does not support .ma/.mb files; scenes are saved in .inca format.
The Solana Alpenglow upgrade went live on Wednesday, marking its largest network upgrade since 2020. It shortens transaction finality from approximately 6 seconds to about 150 milliseconds and reduces validator costs by roughly 60%. The upgrade is achieved by moving validator votes off-chain and packaging them into certificates, with activation planned at the start of epoch 1064.
Eight private keys belonging to Indian certificate authority eMudhra were found to have been generated using extremely small integers (such as 4, 56, and 115) instead of strong random numbers; researchers recovered all of them from public certificates within one second. This vulnerability could lead to forged website certificates. Mainstream browsers have urgently blocked the relevant certificates, but non-browser tools such as curl, Python, Node, Go, and Java may still be affected because they do not check certificate revocation status.
MiniMax officially released the M3.1 Flash model, claiming the previously anonymous model Space Bunny. It ranked first in the Automation-Bench v1.0.6 complex Agent task evaluation and tied for first with Opus 5 on BankerToolBench. Measured results show that combined with MiniMax Code, it completed a 3D campus construction task in 55.4 minutes at a model cost of approximately 6.50 yuan, representing a 54% speedup compared to the DeepSeek V4.1 Flash-1017 combination (121 minutes, approximately 6.52 yuan). Input/output prices are 1 yuan/million tokens and 4 yuan/million tokens respectively, with cache reads costing only 0.1 yuan/million tokens.
Splash 1.4, released by Inco AI, is an open-source local AI inference engine optimized specifically for Apple Silicon. It adds support for the Qwen3.8-Flash-Next model. On a Mac with M5 Pro and 64GB memory, tests showed a maximum response generation speed of 120 tokens/s, short-input generation at 95 tokens/s, and 32K input processing at 1,312 tokens/s. The official claim states it is more than 30% faster than Strata. It supports a context window of up to 1 million tokens, utilizing Metal compute kernels and DFlash 2 speculative decoding. It only supports macOS 26.4+ and chips from M3 onwards. The installation command is brew install incoai/tap/splash. The project uses the Apache 2.0 license.