Tech watch

Unlike voice systems that turn speech into text before responding, Dessa processes audio directly, end to end. Trained only on human conversation, this speech-to-speech model runs full-duplex: it can listen and talk at once, interrupt, or punctuate an exchange with laughter and backchannels. Its demos also show how pauses and intonation convey meaning.

Inference is reported to run faster than real time, with local execution on a MacBook Pro M4 Max or better. This is still a research preview; the team is inviting AI safety researchers to assess the risks of a model that perceives emotion in speech.

Archify is an open-source Agent Skill that helps an AI agent turn a description, question, or repository evidence into an interactive diagram or a concise page with a clear conclusion and sources. It supports ten diagram types, from architecture maps to timelines, and outputs a self-contained HTML file with no runtime dependencies.

Maps can be explored and refined in plain language, without learning a diagram editor. For one-page answers, sources can be pinned to a specific commit; the result can also be exported as a long image or PDF. The skill works with Claude Code, Codex, Cursor, OpenCode, and GitHub Copilot.

This tool adjusts a system font to more closely match a brand font while it loads. It generates the required @font-face rules—including size-adjust, ascent-override, descent-override, and line-gap-override—using a visual preview and adjustable controls.

The approach targets layout shifts caused by differences in font metrics. The resulting CSS can be copied into a project, though the page notes that the tool is not supported in Safari.

When an agent changes a UI, checking a single account cannot cover every role, plan, or feature flag. Matt Rothenberg shows how values scattered across hooks, providers, and layouts can hide overlooked states — such as a Delete button shown to a view-only user.

Making display state an explicit component input lets teams render relevant combinations directly, test them, or compare them in a visual grid. Storybook and Playwright complement this approach, but they do not choose which scenarios to cover; that decision still matters.

Josh W. Comeau reviews ten proposals being considered for Interop 2027: CSS functions, custom border shapes, anchor- and scroll-related queries, reading order, intrinsic-size animations, and grid decorations. Several address practical problems: reading-flow could align keyboard navigation with a responsive layout, while grid decorations could draw separators without adding empty DOM cells.

The proposals are at different stages of maturity. As of October 2026, several are available in Chromium but not yet in Firefox or Safari; inherit() is implemented in no browser. The article shows where progressive enhancement is feasible, with safeguards such as @supports for height transitions.

OGL keeps abstractions to a minimum for WebGL rendering and custom shaders. This dependency-free ES module library shares some conventions with Three.js, but provides fewer features and leaves room to use native WebGL commands directly.

Its Math, Core, and Extras split helps keep integrations modular; the README lists 29 KB minified and gzipped for the full set before tree-shaking. Commented examples cover both a camera-based scene and a fullscreen shader. The low abstraction level makes the code easier to learn and extend, while leaving more WebGL details to the developer.

Chums connects an Apple Watch and iPhone to Claude Code or Codex agents running on a Mac. Its interface makes agent activity visible, shows tool calls, and lets users respond to permission requests or dictate instructions.

The Node.js hub stays on the Mac, but the watch API is exposed through Tailscale Funnel and protected by a token. The project is open source and requires no Chums account, though setup involves building the apps with Xcode and configuring several dependencies—a trade-off between local control and a more demanding installation.

Claude SEO turns Claude Code into an SEO auditing tool: specialist agents assess indexing, performance, content, structured data and visibility in AI search, then produce a prioritized action plan. Reports stay on the machine, but analyzed pages are sent to Claude, as with other Claude Code workflows.

The repository also emphasizes tooling reliability, with automated checks, support for Python 3.11 through 3.14, and safeguards against untrusted web content. Live-data extensions are optional; without Google API keys, Core Web Vitals results are lab estimates.

Cloudflare introduces Clairic, an extraction model available in preview on Workers AI. From text or images, it returns typed values such as dates, numbers, and lists, and can also compute, estimate, or forecast values, with confidence scores and intervals for estimates. Its responses are well-formed, but not necessarily correct.

The API can process multiple fields in parallel and return extracted text with its source offsets. Cloudflare notes limits in the current checkpoint: multi-step arithmetic, estimates without supporting evidence, and detailed forecasting call for caution. The authors recommend leaving complex calculations to code and using confidence scores and intervals to guide review.

The CSS random() function adds visual variation in the browser, without JavaScript or Sass loops. Brecht De Ruyte demonstrates it with subtly tilted letters, cards animated on scroll, and shadows that respond to card rotation. A caching key can keep multiple properties in sync, while sibling-index() calculates the structure of a fanned-out card stack.

These effects are best suited to decorative elements: the article advises against randomizing grids, body text, or interactive targets. With browser support still rolling out, it recommends a stable baseline enhanced through @supports, and stresses respecting prefers-reduced-motion and screen-reader access.

To reduce loading skeletons in v0, the team combines Next.js Partial Prefetching with cache rules suited to public and personalized data. Playwright instant() tests check that priority content is ready at click time, and remain in the codebase as CI regression checks.

Prefetching also responds to user intent: in large grids, details load on hover, focus, or touch rather than for every visible card. That adjustment followed a 2.9× increase in prefetch requests. The overall comparison reports lower daily costs, while cautioning that the change cannot be attributed to this refactor alone.

Vercel Shop pairs a Next.js storefront with Shopify, which remains responsible for the catalog, payments, and checkout. Cache Components serve product data quickly, then stream personalized content; meanwhile, the cart updates before the server responds through an optimistic UI.

The template adds agent-facing options: an opt-in shopping assistant, skills for extending the store, and Markdown served through the Accept header. It offers a concrete starting point for these flows while keeping Shopify as the commerce engine.

Astro 7.4 introduces astro build --timings, breaking down time spent rendering pages, running Vite plugins, processing Markdown, and transforming images. The release example shows how the report exposed sequential network requests and helped cut generation time for blog pages.

The release also separates image services for build and runtime, and adds notFound() for manually composed entrypoints. Parallel prerendering and custom redirect pages remain experimental; the former can increase memory use. One upgrade requirement to check: Astro now requires Node.js 22.22.0 or later.

c15t 3 rebuilds consent management around a shared engine with adapters for Next.js, React, Vue, Nuxt, Astro, Svelte, TanStack Start, and plain HTML sites. Banners can be rendered into server HTML, while a consent manifest avoids waiting on the backend for every request. The reported gains depend on the deployment mode and come from benchmarks published by the project.

The migration preserves choices stored by v2, but changes APIs, packages, and the backend. Codemods handle part of the upgrade; clients and backend must move to v3 together. The release also introduces more strictly validated policy rules and a rewritten self-hosted backend supporting PostgreSQL, MySQL, and SQLite.

React Bits 2.0 announces 216 animated components: 104 have been rebuilt or redesigned, and more than 20 have dropped extra dependencies. Demos add presets and click interactions, while Background Studio, Shape Magic, and Texture Lab have been rebuilt to tune effects live and export the results.

The update also refreshes the documentation and landing page, which is reported to use about five times less JavaScript. The catalog supports expressive interface prototyping; the range of animations still makes it worth assessing their runtime and dependency costs before integrating them into a product.

Mutative lets developers write mutations against a draft to produce a new immutable state, avoiding manual spread copies at every level. It also supports arrays, Set and Map, along with patches and optional freezing.

Its benchmarks report a geometric mean 3.6 times faster than Immer with matched settings, or 6.6 times with each library’s defaults. Results vary by workload; hand-written reducers remain faster in most of the cases shown.