SShortSingh.
Back to feed

VS Code 1.147 Adds Browser Use, Model Routers and Scheduled Requests, Mostly in Preview

0
·1 views

Visual Studio Code 1.147 reached Stable on November 18 and focuses largely on agent features, nearly all of which ship as Experimental or Preview and are off by default. Its headline addition, Browser Use, lets an agent work directly in Chrome or Edge, reusing existing logged-in sessions, though it requires the Copilot browser extension, a local Copilot harness session, and Node.js 22 or later. Model routers, configured through YAML files, use a decision model to send each request to a suitable model, trading cost against quality. Other additions include Send Later for scheduling chat requests, which only runs on machines that are awake, and Drafts for keeping several unsent requests. Admins can also now permit the claude and codex harnesses alongside managed Copilot settings.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Why one writer keeps a wrong-answer notebook to sharpen music listening

A writer describes keeping a two-column notebook that records their wrong guesses about songs alongside the actual answers. They only log misses they were confident about, since lucky guesses and blanks reveal little about their listening habits. The notebook exposed a flawed assumption that most 2000s pop songs were produced by the same four producers. Rather than treating the answer archive as spoilers, the writer uses it as an index to spot repeating artists and production styles. The habit is presented as a simple, free way to track progress in music trivia over time.

0
ProgrammingDEV Community ·

OneFindMe Makes 1,200 Static Guides Searchable by AI Agents Using MCP and NLWeb

The developer behind OneFindMe, a multilingual AliExpress product search, found that an existing MCP server let AI assistants search products but not read the site's roughly 1,200 static guides. They built a lightweight index from each page's own title, description, headings, and FAQ questions, stored as a generated JavaScript module on Cloudflare Pages Functions, with no database or vector store. Keyword search with field weighting and language-aware filtering exposes search_guides and get_guide tools, with get_guide fetching live page content and converting it to Markdown. The site also advertises NLWeb endpoints via Link headers, and a commit to the archive repo added llms.txt files so GitMCP serves current content to coding agents.

0
ProgrammingDEV Community ·

CityEngine rule files keep a university plant model accurate through repeated rebuilds

A university central plant is rebuilt every fifteen years without demolition, so facilities staff needed a model that could absorb ongoing equipment changes. The team built it in CityEngine from campus GIS footprints and an asset table of heights, floor counts, and inspection dates, using a short rule file and Python scripting to export glTF, GLB, OBJ, and JSON with a versioned manifest. The hand-modeled pipe bridge was the only element the rules could not generate, and GenAnywhere's concept image was rejected because its stacks were too short, the bridge was missing, and it had no queryable data. The resulting geometry is shared across a browser viewer, ArcGIS, and a structural consultant, so all users work from the same coordinates. The model is designed to be updated through its attributes when the plant changes again in 2029.