// I was young and needed to ship.

HTML/CSS is Turing complete, so why not compile programs to plain HTML/CSS and runs with zero bytes of JavaScript? zero-js.vercel.app

In Vercel Sandbox when you allow list a host name: - DNS will only resolve for that host name - TLS SNI will only work for that host name - When filtering HTTP(S), the host header must match that host name

These days IP allow-lists are not very useful because large chunks of the internet runs on a relatively small number of IPs. At best it works if you own the destination IP and know it is single use. You need to perform either SNI based allow-listing but if you are not sure whether the destination host might allow domain fronting then the only secure way is to MITM the TLS and also enforce a host header allow list. You probably want to do that anyway because you should enforce stricter rules than allowing the entire host.

Meet Pixel Canary on Vercel AI Gateway. • 𝚜𝚝𝚎𝚊𝚕𝚝𝚑/𝚙𝚒𝚡𝚎𝚕-𝚌𝚊𝚗𝚊𝚛𝚢 • Free in stealth for a limited time • Agentic coding & mobile app dev • Ties GPT-6 Astra on Next.js evals • Prompts may be used for model improvement vercel.com/changelog/pixel-c…

GPT-Sol absolutely mogs on DeepSecBench. Very impressive. Better than Astra at 20% of the price 🤯

In 7 months, the skills.⁠sh registry reached 1 million agent skills and nearly 280 million installs. See what people are teaching agents, and where skills are going ↓ vercel.com/blog/state-of-age…

I was vibing on the beat a little too hard

The AI-based hardening of the entire ecosystem continues 💪 I'm incredibly proud of the leadership of the @nextjs team in surfacing issues and helping drive disclosure through the ecosystem. PSA: If you do any type of image processing directly (not through a framework; and this is very common; O(most apps)), you really must be updating your dependencies at least weekly

🚢 We've been developing a custom S3-backed block storage for the past few months: Drives give you terabytes (!) of storage on Sandboxes, up from 64GB previously, and can be shared on many workloads at once Can't wait to share more about the technical details

Muse, Instinct, OpenClaw, Claude Code… All successful agents have 3 key components: 🧠 Brain → model, harness (logic) 👐 Hands → tools, computer, browser 🗃️ Files → memories, skills, repos The 'easy' way is to throw all these in 1 stateful computer (a Mac Mini) Like, you run 𝚌𝚕𝚊𝚞𝚍𝚎 or 𝚏𝚡 in your mac, you keep it running all day with 𝚌𝚊𝚏𝚏𝚎𝚒𝚗𝚊𝚝𝚎, it has storage, and CLIs and apps installed. But if you want to cost-efficiently run agents in the cloud, you actually start breaking down these parts. 🧠 The harness can run in Fluid compute. To make it reliable across restarts, rollouts, crashes, you make its event log durable using Workflow. 👐 The hands can be a dedicated browser fleet like Browserbase/Kernel, a computer like Sandbox, and even more efficient lightweight tools like just-bash. 🗃️ 🆕 What was missing was a way to also decouple storage. Imagine you want to run a memory consolidation cron job every night ("dreaming"). You can read/write to the files directly without 'booting up' the agent's full computer. Today we're introducing the perfect companion to Sandbox: Drives. We shipped the computer for agents, now we're giving you the 'external disk' you can attach at will. It's early, and we'll be expanding capabilities here quickly. Btw, breaking apart the agent into these independent parts not only optimizes costs in a big way, it also *massively* improves security and auditability. I'd argue you can't even run a secure agent otherwise!

Drives available for Sandboxes on Vercel. Writes are NVMe speed. Reads are block-wise lazy-loaded from blob store and NVMe speed after.

Workflow tools in eve: deterministically orchestrate subagents, add human-in-the-loop, stream progress, and durably execute multi-step work

To all the Agents as a Service startups out there: Get those Enterprise $$$$s.

Have a question? Ask Jev AskJev.net

🤯 @grok Is this true?

Jev is still 🆓 on Vercel for 4 more days

As a weekend project I set out to determine the prime factors of happiness also known as RSA-255. I sent 1000 agents using Astra X high on newly freed up spare GPU capacity from workloads moving from LLMs to Jev and this is what they came up with

Jev is classic low-end disruption. You don't invent Jev is you are a GPU-rich hyperlab

For the same reason that it is more difficult to write the lean proof than it is to validate the lean proof. Is there nuance? Yes. Is the nuance that having the new tool in the toolbox is worthless? No

There are some really wild things you can do with Jev and agent-browser Example: