标签：#open

If you work with external APIs, you know the problem. There is no dev.github.com , no staging.api.companieshouse.gov.uk , no test endpoint for the thing you actually depend on. Production is the only source. So how do you get a development environment without re-fetching everything you already have? The usual answer is to copy: duplicate the warehouse, or keep a separate dev database and sync it periodically. Both are slow, both drift, and both cost storage in proportion to the number of people on the team. Interlace does something else. An environment is not a copy of your data, it is a set of views over it. Fingerprints first Every model gets a fingerprint: a hash of its canonical SQL — or its Python source — together with its strategy configuration and its upstream fingerprints. A build writes an immutable physical table named after that fingerprint. interlace__main.orders__a1b2c3 That table never changes. If the model's definition changes, the new version gets a new fingerprint and a new table, and the old one stays exactly where it is. An environment is then just a set of views pointing at fingerprinted tables. Production is the unprefixed namespace; every other environment prefixes its schema. Environment View for main.orders prod main.orders dev dev__main.orders pr-142 pr-142__main.orders Consumers and BI tools connect to main.orders and never learn that a fingerprint exists. There is no environment list to configure, either — an environment exists once something has been promoted to it. Why the sandbox is free Here is where the re-fetching problem disappears. Applying to a sandbox does not rebuild models whose fingerprint already exists. It points the sandbox's views at the tables production already built. interlace apply --env dev Change one model out of forty and the sandbox builds one model. The other thirty-nine are reused — not copied, reused, the same physical tables production is reading. The expensive source extract that ran this morning is the table

You added an MCP server to your AI assistant. Did you check what it can touch? MCP servers give your AI assistant new abilities: read your filesystem, query your database, call an API, run a shell command. That is the whole point of them. It is also the whole point of the risk. The permission question nobody asks When you install a normal browser extension, you at least see a permission prompt. When you add an MCP server to your AI coding assistant, you usually do not. You add a config entry, restart, and the assistant now has whatever access that server exposes. Most people never read the server's source to see what that actually is. This matters more with AI-built or AI-suggested MCP servers specifically. If the assistant wrote the server for you, or you copied one from a repo you have not read closely, you have no independent confirmation of what it does versus what its description says it does. What tends to go wrong Three patterns show up repeatedly: A server meant to read files ends up with write access too, because the broader permission was easier to implement and nobody scoped it down. A server that talks to an external API embeds a credential directly in its config or source, so anyone who can read the server's files can read the key. A server built for local development gets pointed at a production database or production credentials once it "works," without a second look at what commands it now accepts. None of this requires anyone to be careless in an obvious way. It is the same gap as any fast-shipped code: the server works, so it ships, and the access-scoping step that would normally happen in review gets skipped because there was no review. A practical check before you trust an MCP server Before you add an MCP server to a live setup, or before you point an existing one at anything real: Read what the server can actually do, not just its stated purpose. Check the tool definitions it exposes, not the README. Check where its credentials live. A server th

Cloudflare launched agent tracing, adding spans for agent invocations, model calls, tool runs, and approvals to existing Workers traces. Sessions replay turn by turn, though the docs warn traces are not lossless and payloads may be truncated. Payload recording defaults differ by framework, and from October 1, 2026 every span counts as a billable event. By Steef-Jan Wiggers

Most of the industrial AI you have seen is a retrieval pipeline with a chat box on it. Chunk the manuals, embed them, stuff the top matches into a context window, let the model talk. It demos well. It falls apart the first time somebody asks a question where the answer depends on what a machine is doing right now. We are an applied research lab called Nodeblue, and the system we build is called Nexus. This is a writeup of the architectural decision at the center of it, which is that the language model is the smallest and least interesting part. The failure that set the design Here is the test that made the decision for us. Industrial control programs live in two places. There is the project archive, which is the file in source control, and there is the program actually running in the processor. Those drift, constantly, because engineers go online to fix a timer during a downtime event and do not always upload the change back. Anyone who has worked on a plant floor knows this. It is Tuesday. We took a real production controller with exactly that situation, a live edit present in the processor and absent from the archive, and asked eleven frontier models which version of the routine was executing. We gave them the export, the docs, the context, everything a careful human would get. All eleven answered confidently. All eleven were wrong. Not garbled, not obviously broken. They read the export correctly, described the rung correctly, and then told us the archived version was running, because the archived version was the only version they had ever seen. Then we put the same eleven models on top of our engine and asked again. All eleven got it right, cited to the rung. The models did not improve. They got access to a fact that lives in a processor rather than in a corpus. That is the entire lesson, and it generalizes past our domain: when a model is asked something it structurally cannot know, it does not abstain, it produces the most probable sentence. In a domain where

I've evaluated every open-source WhatsApp bot framework on GitHub. They all share the same fatal flaw. The Problem Nobody Talks About Most WhatsApp bot frameworks are glorified API wrappers. They handle message transport — receiving a text, routing it somewhere, sending a reply — and that's it. The "intelligence" layer is left entirely to you. You get a pipe. You get a webhook. You get some session management. And then you're on your own. The frameworks that do add AI make a different mistake: they duct-tape GPT onto the messaging pipe and call it "AI-powered." The pattern is always the same: receive message → append to conversation history → call openai.chat.completions.create() → send reply. It's generic. It's stateless in any meaningful business sense. It doesn't know what industry it's serving, what data it has access to, or what actions it's actually allowed to take. Here's the part that breaks me: none of these frameworks understand that a restaurant needs different tools than a law firm . A restaurant needs to check table availability, query allergens, create reservations, and handle cancellations. A law firm needs to schedule consultations, check document status, route inquiries by practice area. These are not the same problem. Treating them as "just chat" is the core architectural failure of every framework I've seen. And then there's the "enterprise" tier: Twilio Flex, Intercom, Freshchat. These charge $500–$2,000/month for what is fundamentally a prompt and a webhook wrapped in a dashboard. They're selling you infrastructure and calling it intelligence. The underlying model doesn't know your business. It can't execute actions in your systems. It's an expensive illusion. What's Actually Needed The shift that matters isn't from "no AI" to "has AI." It's from generic chat to domain-specific function calling . This is not a subtle distinction. Here's what a properly architected tool dispatcher looks like versus what everyone else ships: // Each vertical gets

For a long time, my mental model of observability was this: you import an SDK, sprinkle some calls through your code, each call fires off data to a server somewhere, and a dashboard reads it back. A logging system with extra steps. That model is wrong in a specific, interesting way. And I couldn't see how it was wrong until I stopped looking at the dashboards and started looking at what actually gets emitted, and how. The seductive wrong model The wrong model is seductive because the plumbing really does look identical. Logging: emit, store, search. Observability: emit, store, query. Same loop, right? So my working theory became: observability is logging plus some fancy logic to analyze the logs. Close. But no. The difference isn't in the analysis. It's in the emission — and it splits into three mechanisms that have almost nothing in common with each other. Descent one: metrics aren't events at all A metric is not a record you write. It's a number sitting in your app's memory . requests_total . increment () // 1, 2, 3... request_duration . record ( 0.23 ) // adds to a histogram Nothing is sent when this line runs. The number just changes in RAM. Periodically — every 15 seconds, say — either a backend scrapes an endpoint your app exposes, or a collector ships the current values out. That's why metrics are absurdly cheap: a million requests is one counter reading "1,000,000", not a million records. You could never reconstruct a clean p99 latency graph by parsing log text. The histogram was built for it at write time. And the stateless-container objection answers itself: the in-memory counter is disposable. Each instance flushes to the backend on a schedule (on serverless, a sidecar collector even does a final flush at shutdown), and the backend sums across instances. The durable truth never lived in your app. Descent two: logs are the familiar part Logs work exactly the way I always assumed everything worked: an event, written out, shipped, searched. The only upgrade

Qwen 3.8 arrived as two different releases with two different licences, and only one of them is something you can put on a card you own. The 2.4 trillion parameter A95B opened up on 12 August under Alibaba's own qwen3.8-max terms. The one that matters for local work is Qwen 3.8 27B , whose safetensors went up on 13 August at 08:23 UTC with an Apache 2.0 LICENSE file following the next morning. Both dates are off the Hugging Face commit log, not a launch post. Here is the practical picture: what it needs, why its long context is unusually cheap, and the one setting that makes people think they downloaded a broken quant. The shape of the model decides everything 27B dense parameters across 64 layers, hidden size 5120. The interesting part is in config.json , where layer_types reads 48 linear attention layers and 16 full attention layers , alternating three to one ( full_attention_interval: 4 ). Only those 16 layers keep a KV cache. The rest of the shape: 24 attention heads with head_dim 256 and 4 KV heads , a 248,320 token vocabulary, and max_position_embeddings of 262,144 . It is a native vision language model, so images and video go in without a wrapper, and the ggml-org pack also ships a multi token prediction head as a separate file. The numbers Sizes below are the file sizes Hugging Face reports for unsloth/Qwen3.8-27B-GGUF , read on 14 August 2026. Packs differ by a few hundred megabytes, so check the repo you actually pull from. lmstudio-community has Q4_K_M at 16.8 GB and ggml-org at 19.0 GB for the same nominal quant. Quant Size on disk Realistic home UD-IQ2_XXS 9.0 GB 12 GB cards, visible quality cost UD-Q2_K_XL 10.7 GB 12 GB cards, almost no context left UD-Q3_K_XL 13.4 GB 16 GB cards Q3_K_M 13.8 GB 16 GB cards IQ4_XS 15.7 GB largest quant that stays whole on 16 GB Q4_K_M (sweet spot) 17.1 GB 24 GB cards Q5_K_M 19.8 GB 24 GB, less context headroom Q6_K 22.9 GB 24 GB barely, or 32 GB Q8_0 29.0 GB 32 GB or a two card split BF16 (from ggml-org ) 53.8 GB server

A research announcement from a working multi-agent operation. Full paper to follow. A word first, on spirit. I am not a scientist, and none of this was done in a laboratory. It came out of my own work, something I built to get a job done and then could not stop looking at. Nothing here is a knock on the companies whose tools I use. What they have built is remarkable, and it is getting better by the day. I am not testing their systems to find fault. I am testing them to learn how each one handles the persistence of memory, personality, and self across sessions, in a single-agent and multi-agent design. If you build with these tools, the next paragraph is familiar ground. If you don't, it is the ground everything else here stands on. Here is one example of how an AI agent currently works by default and what the system I built changes. Every conversation runs inside a context window, a session with a token limit, billed against your online subscription account. At the start of a session three files load: the root file, a room file that tells the agent who it is, and a memory file which is capped at 25,000 characters, or 200 lines, a limited index. All of them load automatically. The memory file is really the only constant reference the agent has to past sessions, and it provides pointers to a folder of one-line notes, but no rule or hook makes it read the notes. Going deeper is left to the model, and often it doesn’t. The notes sit referenced but unread while the agent answers from what’s already in front of it in the current session. After that the model, the raw AI engine, keeps nothing between turns; each turn the model re-reads the whole conversation from the top and rebuilds its understanding from that. The software that holds this conversation and runs the model’s tools is the harness, and every commercially available AI system has one. As the session fills, the platform summarizes it, and the agent understands less, a kind of attenuation, the way an audio or vid

I kept paying a monthly subscription for a cloud coding assistant while a 16GB M4 Mac mini sat on my desk idling most of the day. So I ran the obvious experiment: can a 16GB Mac mini run a coding assistant entirely offline — no code leaving the machine, no subscription — and is it actually usable for real work? Short answer: yes, with one hard constraint (RAM) and one soft one (context length). This article is the written version of the video above, with every command, config file, and benchmark number so you can reproduce it. Table of contents Why bother running locally The hardware constraint nobody mentions Step 1: Install Ollama Step 2: Pick a model that fits in 16GB Step 3: Run and verify Step 4: Wire it into VS Code Step 5: Tune Ollama for a 16GB box Benchmarks What it does well, what it doesn't Should you cancel Copilot? Why bother running locally Three reasons, in the order that actually mattered to me: Privacy. Client code, internal repos, anything under NDA — none of it leaves the machine. This is the one thing a hosted assistant cannot offer you at any price tier. Cost. A coding assistant subscription is roughly $100–240/yr depending on tier. The Mac mini was already bought. Offline. Flights, bad hotel wifi, coffee shop dead zones. The assistant just works. The reason not to: raw capability. The frontier hosted models are better at large multi-file reasoning, and it isn't close. More on that below. The hardware constraint nobody mentions On Apple Silicon, the GPU and CPU share one pool of unified memory. A model has to fit in that pool alongside macOS, your browser, VS Code, and whatever containers you're running . On a 16GB machine, macOS + a normal dev environment eats 6–8GB before you've loaded anything. That leaves you roughly 7–9GB of realistic headroom for the model. This single number determines everything else, and it's why "just run the 30B model" advice from people on 64GB machines doesn't transfer. By default macOS allows the GPU to use about 7

What happens when you stop picking a stack and let each protocol pick one for you. Every engineering team has a stack. Ours has seven, and we did not decide on any of them. Nevai is a self-hosted, end-to-end encrypted workspace built on Matrix. Part of it is a set of bridges — 25 of them — connecting Discord, Telegram, WhatsApp, Signal, iMessage, Messenger, Instagram, Slack, Google Chat, LINE, WeChat, KakaoTalk, Skype, GroupMe, SMS, email, IRC, XMPP, Zulip, Mattermost, Revolt, Mumble, QQ, X and LinkedIn into one place. We started out intending to standardise. We ended up with this: Language Bridges Go 12 TypeScript 5 Python 4 JavaScript 1 Kotlin 1 PLpgSQL 1 Slice 1 Nobody sat in a room and chose that distribution. It is what you get when the protocol decides. Go wins where the protocol was reverse-engineered WhatsApp, Signal, iMessage, Messenger, Instagram, Telegram, X, WeChat, QQ, Skype, LinkedIn, email. Twelve bridges, and the reason is the same every time: the mature libraries for those protocols are written in Go. That is not a claim about Go being a better language. It is a claim about where a decade of reverse-engineering effort happens to live. If you want to speak WhatsApp's protocol without running a browser session, you use what exists, and what exists is Go. Look at what leaks in around the edges and the picture gets sharper: Signal is 86% Go and 13% C — the C is libsignal, and you do not reimplement libsignal. iMessage is 96% Go and 3% Objective-C — because iMessage runs on macOS, and at some point you have to talk to the operating system in its own language. Those percentages are the honest part. A bridge is mostly your code and a small amount of somebody else's, and the small amount is usually the part that matters most. Python wins where the API is boring Google Chat, Zulip, KakaoTalk, LINE. Documented HTTP APIs, JSON in and JSON out, no protocol archaeology required. There is no performance argument here. These bridges are not throughput-bound; they

Imagine this: You’re drafting a long email, writing a report, or responding to a wave of Slack messages. Instead of hunching over your keyboard and typing at 40 words per minute, you simply hold down Ctrl + Space , speak your thoughts at 150+ words per minute, and release the keys. Instantly, clean, perfectly punctuated, polished text appears right where your cursor is. Meet Vacanam — a free, 100% private, offline voice typing tool built for Windows 10 & 11. 😫 Why Most Voice Typing Tools Are Frustrating If you’ve ever tried built-in dictation tools or commercial transcription services, you’ve likely run into the same annoyances: They Send Your Voice to the Cloud : Many tools stream your microphone audio to remote servers. If you work with sensitive emails, client data, or private thoughts, that’s an immediate dealbreaker. They Require an Internet Connection : Try dictating on an airplane, during spotty Wi-Fi, or in a secure offline room — they simply refuse to work. Punctuation is a Headache : You have to awkwardly say things like "Hello comma how are you question mark" just to get a basic sentence right. Subscription Fatigue : Most good dictation apps charge $10 to $30 every single month. We built Vacanam (वचनम् — Sanskrit for Voice & Speech ) to fix all of this once and for all. 🌟 The Superpowers: What Makes Vacanam Different? 1. 🎙️ Works in Every Single Windows App Vacanam doesn’t trap you inside a special recording window. It works universally: Productivity & Docs : Microsoft Word, Google Docs, Notion, Obsidian, OneNote Communication : Slack, Microsoft Teams, WhatsApp Desktop, Discord, Outlook, Gmail Browsers & Editors : Chrome, Edge, Firefox, Notepad, VS Code, Terminals Just click into any text box, hold Ctrl + Space, speak, and let go. 2. 🪄 Automatic AI Polish (No More "Ums" or Missing Commas) When we talk, we hesitate, say "um" , repeat words, and forget punctuation. Vacanam features an optional Built-in AI Assistant that runs silently on your computer: Remov

Tim O’Reilly built a publishing empire that AI is helping to destroy. Yet he loves AI—as long as it’s open source.

US groups release cheaper models after new challenges to their trillion-dollar ambitions.

GitHub热门项目 | ML-powered manga translator, written in Rust. | Stars: 5,136 | 109 stars this week | 语言: Rust

GitHub热门项目 | 🚀 通用 AI IDE 账号管理工具：支持 Antigravity / Codex / GitHub Copilot / Windsurf / Kiro / Cursor / Gemini-cli / CodeBuddy，多账号切换、配额监控、自动唤醒与多开实例管理。 🚀 Universal AI IDE account manager for Antigravity / Codex / GitHub Copilot / Windsurf / Kiro / Cursor / Gemini-cli / CodeBuddy, with multi-account switching, quota monitoring, wake-up automation, and multi-insta | Stars: 15,751 | 75 stars today | 语言: Rust

GitHub热门项目 | Collection of effects for React: Border beam, Liquid Gooey | Stars: 1,921 | 54 stars today | 语言: TypeScript

GitHub热门项目 | 🙌 OpenHands: AI-Driven Development | Stars: 84,010 | 120 stars today | 语言: TypeScript

GitHub热门项目 | 免费无广告的追剧资源指南，人工精选资源、每天检测资源有效性。收录在线影视、影视APP、网盘搜索、磁力BT、字幕、TVBox / 影视仓空壳软件/配置地址、IPTV直播源、会员拼团、影视相关开源项目。开源，社区共同维护。 | Stars: 5,766 | 71 stars today | 语言: JavaScript

GitHub热门项目 | 🎬 Fully automated YouTube channel management with AI agents. Creates, optimizes & publishes videos 24/7. Works with FREE Gemini API or OpenAI. No coding required! | Stars: 1,937 | 118 stars today | 语言: JavaScript

GitHub热门项目 | An open-source, GPU-accelerated physics simulation engine built upon NVIDIA Warp, specifically targeting roboticists and simulation researchers. | Stars: 5,341 | 15 stars today | 语言: Python