#browserai
Jeyzma is like @typesafeai.bsky.social minus $870 million plus @golang.org & @tinygo.org & @hf.co llama.cpp

Go make a decision!

Already available right now in your browser.

jeyzma.com

#jev #decisionAI #browserAI #llamacpp #systemOne
Jeyzma - Go make a decision
Go make a decision. System One decisions, 100% local in your browser.
jeyzma.com
October 10, 2026 at 8:52 AM
Jeyzma also keeping "browser AI" hot, updated with new models Julia-1 & Laya. Plus the latest high-performance WebGPU builds! Decision support completely local all written in @golang.org with @tinygo.org

jeyzma.com

#golang #tinygo #localAI #browserAI
Jeyzma - Go with your own intelligence
Go with your own intelligence. System One decisions, 100% local in your browser.
jeyzma.com
October 7, 2026 at 10:54 AM
Shopify just let browser AI agents finish the checkout for you—no extra steps, just smooth cart-to‑pay via WebMCP and Shop Pay. Your next purchase could be AI‑driven. Curious? Dive in. #ShopifyAI #BrowserAI #WebMCP

🔗 aidailypost.com/news/shopify...
September 28, 2026 at 10:59 PM
BragJack shows how a single malicious browser extension can hijack AI assistants in Chrome, Edge, Opera Neon, Comet, and Claude in Chrome, leading to two CVEs and vendor bug bounties. #BragJack #BrowserAI #Chrome
BragJack Attacks Hijack AI Browser Agents Through Malicious Extensions
Security researcher Gal Weizman of Forever Security revealed BragJack, a new attack technique that can hijack AI assistants inside popular browsers through a single malicious extension. The findings affected Google Chrome's Gemini Live, Perplexity Comet, Microsoft Edge, Opera Neon, and Claude in Chrome, and led to two CVEs and bug bounty rewards from multiple vendors. #BragJack #GalWeizman #ForeverSecurity #GoogleChrome #GeminiLive #PerplexityComet #MicrosoftEdge #OperaNeon #ClaudeinChrome #CVE-2026-0628 #CVE-2026-55945
www.hendryadrian.com
September 19, 2026 at 7:15 PM
用 Vercel AI SDK 玩轉 Chrome Built-in AI
Google 官方關於 Chrome Built-in AI 與 Vercel AI SDK 的文件寫了一大堆,但看完會發現官方範例只教你怎麼印出 Hello World,好多坑都沒有交代。 如果模型還沒下載完該怎麼辦?瀏覽器不支援時怎麼辦?Tool Calling 的介面要怎麼安全的接上? 附上在 Chrome上的一些踩坑小筆記,幫你想在前端加上本地 LLM 時少走彎路(雖然彎路的都記得比較久XD)! ## 官方範例 Google 官方有兩篇基礎教學,繁體中文版都寫得清楚: * Prompt API 搭配 Vercel AI SDK * 使用 AI Elements 打造 UI(含 demo) 看完你大概可以了解,可以用`browserAI()` 把 model 丟進 `generateText`、`streamText`、`Output.object`,大概 30 行就能跑,但官方範例大多著重在「API 長什麼樣」... ## @browser-ai/core `@browser-ai/core` 是 Chrome Built-in AI 接到 AI SDK 的 provider,特點是同時支援 AI SDK v5-7,且跨三個本地引擎(Prompt API、Transformers.js、WebLLM )共用同一套 API,目前在「瀏覽器本地推理」這個領域已經是最主要的Tool。 作者是 jakobhoeg/browser-ai的 Jakob Hoeg Mørk,2025 年開始維護,進過 Chrome for Developers 的贊助名單,現在掛在 Vercel OSS Program 底下。 import { generateText } from "ai"; import { browserAI } from "@browser-ai/core"; const result = await generateText({ model: browserAI(), prompt: "Invent a new holiday.", }); 當「模型還沒下載」的狀況發生時,`availability()` 會回 `unavailable`(瀏覽器不支援)或 `downloadable`(可以下載),配合 `createSessionWithProgress` 我們就能拿到下載百分比,這對于前端無論開發什麼樣的應用,都是給使用者第一層的UX,大部分的首次下載要吞好幾 GB的等待時間。 ## next-hybrid 在 `jakobhoeg/browser-ai` 同一個 repo 底下的 `examples/next-hybrid` 也滿值得一看的,它比官方範例完整太多,而且目前都是完整的可demo範例,建議可以直接當 template 複製,他也像我們之前測試transformer/webgpu 一樣,同時接了三種引擎(browser / transformers-js / web-llm)也在UI上做了下拉選單可以即時切換,讓使用者可以同一份 `streamText` 程式碼,本地模型、WebGPU 模型、雲端模型切換成整合或是測試。 有兩個實作細節滿值得整合前端應用的參考的: * 自訂 ChatTransport 接 `useChat` 用 `ClientSideChatTransport` 把 React 的 `useChat` 綁到本地模型來讓整個聊天流程跑在客戶端,連 server 都不用起,萬一有一天你不想用本地模型了,換回 `DefaultChatTransport` 就切到 server API,前端程式碼一行不用改。 * tool calling 的完整流程 它示範了 `webSearch`(用 Exa API)和 `getCurrentTime` 兩個 tool,加上 `toolApproval` 的確認 UI,如果模型想動用外部 API 前,會先跳出一個確認框讓使用者同意確認,這對安全敏感的醫療或金融企業場景,可以配合紀錄一些符合法規面需求的Audit Log ## domainstack.io jakejarvis/domainstack.io 是目前找得到最接近 production 的案例,而且是少數用 `@browser-ai/core` 的第三方專案,它做的是網域名稱查詢工具,雖然串接 AI 只是其中一小部分功能,但架構完整到可以拿來當作範本。 跟我們前面提到的場域類似,核心設計是三種模式自動切換:`local`(本機)、`cloud`(雲端)、`auto`(本機不可用就 fallback),`useBrowserAI` hook 會先查 `availability()`,如果瀏覽器不支援就默默的切到雲端 API,對於使用者完全無感。另外它示範了讓本地模型呼叫 tRPC 的 tool(WHOIS、DNS、SSL 查詢),也證明本地 LLM 做 agentic workflow 是可行的,不是只能聊天。 ## 實測小小筆記 1. API 只能在安全環境(Secure Context)運行 如果你在 about:blank 測試,window.LanguageModel 會直接回傳 undefined,改到 http://localhost 或 HTTPS 環境下,API 才會正常。 2. 模型下載必須由「真實使用者互動」觸發 直接在程式碼初始化呼叫 LanguageModel.create() 會直接報錯:NotAllowedError: Requires a user gesture,這代表不能在頁面載入時自動背景下載,必須由使用者點擊按鈕等主動觸發。因此,如果你用 Headless 測試,必須透過 CDP(Chrome DevTools Protocol)發送 Input.dispatchMouseEvent 來模擬真實的滑鼠點擊事件才可以正常的執行。 3. 硬體硬性門檻 在 chrome://on-device-internals 的 Broker State 系統紀錄中,明確標示需要 20480 MiB required,測試時我的 MBA 原本只剩 4.7GB,模型下載停住,直到清出 23GB 空間後,下載狀態才順利轉綠開始執行。 總結來說,Chrome Built-in AI 的核心優勢不在於頂尖的算力或模型表現,而是在於零 API Key 成本、零代管費用,以及資料絕對不出裝置的隱私安全,再搭配 Vercel AI SDK 後,本地模型與雲端 LLM 的切換只需要一行 Provider 設定就可以搞定。 如果你正準備踹踹,建議路線如下: 快速驗證:用 browserAI() + generateText,30 行內快速跑通。 完整架構學習:直接 Fork next-hybrid 當專案骨架。 生產環境:參考 domainstack.io 的自動 Fallback 策略與 Tool Calling 設計。 目前整個瀏覽器本地 AI 生態還處於早期階段,大部分開發者都還在摸索,趁現在掌握這些工具與解決方案,就能為你的前端專案搶先整合本地 AI 的優勢!
dev.to
August 29, 2026 at 1:09 AM
"Browser AI" is often used as shorthand for "private." It is not that simple. I checked what stays on your device, what downloads, and when network calls or analytics matter on Access Free Tools.

accessfreetools.com/blog/browser...

#BrowserAI #Privacy
Browser AI vs Local AI Privacy: What Stays on Your Device?
Brendan Chambers explains how browser AI, local AI, cloud model calls, downloads, and analytics affect privacy, using real Access Free Tools examples.
accessfreetools.com
August 8, 2026 at 7:10 AM
One of the biggest productivity upgrades I've made wasn't another app—it was improving my browser workflow with AI.

#LearnAI #AICommunity #BrowserAI #AIEducation #PracticalAI #AIWorkflow #OpenKnowledge #FreeAITools #Productivity #DigitalSkills
July 30, 2026 at 6:02 PM
🚀 JavaScript entwickelt sich weiter – und AI gleich mit.

Bei den JavaScript & Angular Days erwarten euch praxisnahe Workshops zu Browser-AI, TypeScript und modernen Developer-Workflows.

#JavaScript #Angular #TypeScript #WebDevelopment #BrowserAI @sirpepe.bsky.social @entwicklerde.bsky.social
July 17, 2026 at 12:31 PM
KI im Browser 2026: Modelle mit WebGPU und LiteRT.js on-device ausführen

KI-Modelle direkt im Browser auszuführen ist endlich schnell genug, um real zu sein. Ein praktischer Leitfaden 2026 für On-Device-Inferenz mit WebGPU und Googles neuem LiteRT.js — was sich…

#browserai #webgpu #litertjs
KI im Browser 2026: Modelle mit WebGPU und LiteRT.js on-device ausführen
KI-Modelle direkt im Browser auszuführen ist endlich schnell genug, um real zu sein. Ein praktischer Leitfaden 2026 für On-Device-Inferenz mit WebGPU und Googles neuem LiteRT.js — was sich geändert hat, wie es funktioniert und wann man es einsetzt.
www.alekseialeinikov.com
July 10, 2026 at 1:27 PM
Browser AI in 2026: Running Models On-Device with WebGPU and LiteRT.js

Running AI models directly in the browser is finally fast enough to be real. A practical 2026 guide to on-device inference with WebGPU and Google's new LiteRT.js runtime — what changed, how i…

#browserai #webgpu #litertjs
Browser AI in 2026: Running Models On-Device with WebGPU and LiteRT.js
Running AI models directly in the browser is finally fast enough to be real. A practical 2026 guide to on-device inference with WebGPU and Google's new LiteRT.js runtime — what changed, how it works, and when to reach for it.
www.alekseialeinikov.com
July 10, 2026 at 1:27 PM
Transformers.js experiments with Cross-Origin Storage API

#TransformersJs #WebStandards #BrowserAi
June 23, 2026 at 6:04 PM
Cloud-free AI in the browser at #MLcon San Diego 🔮

Maximiliano Firtman showed how LLMs, image and speech models can run locally in web apps — reducing backend costs and privacy risks.

NY: https://tinyurl.com/km88wja2
Berlin: https://tinyurl.com/59u4jun9

#BrowserAI #WebLLM
June 5, 2026 at 10:30 AM
Chrome just rolled out repeatable AI ‘Skills’—save your Gemini prompts and reuse them across tabs. No more re‑typing, just smarter browsing. Curious how this changes web workflow? Dive in. #ChromeAI #Gemini #BrowserAI

🔗 aidailypost.com/news/chrome-...
April 14, 2026 at 7:03 PM
What if AI runs directly inside the browser?

No external APIs.
No data leaving the device.

Privacy-first AI is becoming real.

Explore browser-based AI at Voxxed Days Zürich.

📅 vdz26.voxxeddays.ch/

#BrowserAI #WebDevelopment #VDZ26
March 22, 2026 at 1:59 PM
Big shake‑up at Amazon AGI Lab: David Luan exits after steering the Nova Act AI agents and browser‑AI push. What’s next for web automation and true AGI? Dive into the details. #NovaAct #AGI #BrowserAI

🔗 aidailypost.com/news/amazon-...
February 25, 2026 at 3:42 PM
February 20, 2026 at 11:36 AM
I was speaking to this 2.5 years ago. Now there is a protocol for it, #MCP. #Anthropic #WebMCP This is speaking of the next step where communication is #BrowserAI via #WebMCP to #WebServerAI. So back then was still considering further out than now. The Future baby. That's where I live.
February 14, 2026 at 1:19 AM
January 24, 2026 at 11:41 AM
January 15, 2026 at 6:42 PM
You don’t need cloud GPUs to build with AI.

WebLLM, Free LLM Playground, BrowserAI, AgentLLM, and Genspark.ai let you run and compare models right in your browser.

No servers. No installs. No credit card.

What’s stopping you from trying browser-first AI this week?
December 14, 2025 at 5:57 AM