VS Code 1.145 Is Out: Remote Agent Previews and Cron Automations

Visual Studio Code 1.145 shipped as a stable release on November 4, 2026, and it's the first build where Microsoft lets you preview a remote agent's running app, mix different agent harnesses inside one chat session, and schedule local automations with raw cron expressions. Two patch releases, 1.145.1 and 1.145.2, followed within days to fix bugs reported against the initial build. The release also confirms that the Copilot harness, built on the Copilot SDK, is now the default for every new chat session in the editor, while three agent tools are being retired from the older Local harness entirely.

Here's everything that changed, what's still experimental, and how to get the update.

VS Code has shipped a numbered update roughly every month this year, and 1.145 keeps that cadence — arriving almost exactly four weeks after 1.144 and continuing the same broad push toward agent-centric workflows rather than purely editor-level polish. Most of what's new this time sits behind a setting flag rather than being switched on for everyone, which is typical for Microsoft's rollout style: ship a preview, collect feedback through GitHub issues for a release or two, then flip the default once the feature has stabilized.

What's new in VS Code 1.145, at a glance

The release notes group the update into agent, chat, MCP, editor, and enterprise changes. The table below summarizes the features most developers will notice first.

Remote agent sessions get a browser and canvases

When an agent session runs on a remote agent host, the integrated browser inside the Agents window now loads pages through that remote machine instead of your local one. In practice, that means if an agent starts a dev server on a remote box, you can open the running app right inside the Agents window to check the agent's work and leave feedback, the same way you would with a fully local session. Microsoft frames this as catching remote sessions up to a capability that remote workspaces already had.

The feature sits behind the workbench.browser.enableRemoteProxy setting and is explicitly labeled Preview, so expect rough edges. A companion feature, remote canvases, extends the same remote-proxy plumbing to interactive canvases, which previously only worked when everything ran on your own machine. Canvases hosted remotely need both workbench.browser.enableRemoteProxy and chat.canvases.enabled turned on. Teams running agents in the cloud or on beefier remote hardware, rather than a laptop, are the obvious audience here, and it continues a pattern from last week's 1.144 release, which pushed cloud sessions and Copilot Fusion further into general availability.

Mixing agent harnesses inside one chat session

This is arguably the headline feature of the release. Nested sessions in VS Code let you keep related conversations grouped under one parent chat. Until now, every nested session had to use the same harness as its parent — if you wanted a second opinion from a different agent, you had to start an entirely separate session and lose that grouping.

In 1.145, a nested session can run on a different harness than its parent. The example Microsoft gives: ask Claude to review a plan that a Copilot chat just wrote, or hand Codex the same task, side by side, inside the same session. Each nested chat keeps its own model, agent, and permission level based on its harness, and switching one doesn't touch the others. When a chat uses a different harness than its session, it names that harness on hover. The feature is gated behind chat.agentHost.multipleChatProviders.enabled and marked Preview, and Microsoft is explicitly soliciting bug reports through the VS Code issue tracker on GitHub.

It's a notable move in a crowded field where developers increasingly compare editors on exactly this kind of multi-model flexibility — a point worth keeping in mind next to how VS Code stacks up against Cursor and Windsurf on agent workflows.

Cron schedules arrive for local automations

VS Code's automations feature lets you schedule agent tasks to run on their own rather than kicking them off manually. Previously, scheduling options were limited to a handful of built-in presets (hourly, daily, and similar). The 1.145 release adds a raw Cron option: select it in the schedule dialog and enter a standard five-field cron expression when the presets aren't precise enough. The dialog validates the expression as you type and describes, in plain language, the resulting schedule. Cron scheduling is restricted to local automations and isn't available for cloud automations.

Two smaller automation changes ship alongside it: you can now pick a workspace custom agent when creating or editing a local automation, and disabled automations collapse into their own section so they stop cluttering the run history of active ones.

Chat gets verbosity controls, inline attachments, and faster reloads

Several chat-window changes aim at long-running agent conversations. A new, experimental chat.persistentProgressVerbosity setting adds a View Mode picker in the Chat view's overflow menu with three options: Conversation (reasoning and tool calls collapse into sections between response blocks), Verbose (every tool call stays expanded with no scroll limit), and Compact, which is the default and previews tool calls live before collapsing finished ones into summary rows. Switching modes only changes how the transcript displays — it doesn't change what context the agent has or the work it performs.

Inline chat attachments, gated behind chat.inlineAttachments.enabled, move file and image reference pills into the prompt text itself instead of a separate row above the input; it's on by default in Insiders but off by default in Stable for now. Separately, chat can now pull in files from anywhere on your computer, not just the open workspace, through the standard OS file picker — useful for a screenshot or mockup sitting on the desktop. And VS Code trimmed how it re-renders long restored chats, reusing existing reasoning titles and deferring terminal-output rendering until you actually expand a tool call, which should make reopening lengthy Copilot harness conversations noticeably snappier.

The Copilot harness is now default, and three tools leave the Local harness

Microsoft reiterates that the Copilot harness, built on the shared Copilot SDK, continues to be the default for new chat sessions in VS Code, with a dedicated transition guide for teams still moving workflows off the older Local harness. That migration has teeth in this release: the Install Extension (#installExtension), Run VS Code Command (#runCommand), and Create New Workspace (#newWorkspace, formerly #new) agent tools are all removed from the Local harness, along with the /new slash command and the "Generate New Workspace" Welcome page link. Custom agents or prompt files that still reference those tools need to drop them. If your organization hasn't moved its Copilot policies around model deprecations, it's also worth checking the desk's coverage of upcoming GitHub Copilot model deprecations, since harness and model changes tend to land in the same release windows.

MCP: managed allowlists now override registry-only restrictions

For organizations running their own Model Context Protocol server registries, 1.145 fixes a gap that forced every approved server to also be published to the registry. The enterprise-managed allowedMcpServers setting is now authoritative for MCP servers that VS Code itself starts: a server matching the allowlist can run even if it was never added to the registry, and a registry-listed server that doesn't match the allowlist gets blocked. An empty allowlist ([]) blocks everything; if the setting is absent entirely, the old registry-only behavior still applies. User and workspace-level allowlists still can't override the enterprise one, and a deniedMcpServers entry or MCP being switched off entirely still wins over any allowlist. Admins configure this through GitHub's managed settings for enterprise Copilot policies. Separately, MCP servers that expose interactive "MCP apps" can now register tools the model can call directly in its next turn — Microsoft points to a community ShaderToy server as an example of a server letting a model iterate on shader code live.

Smaller editor fixes: file exclusions and Mermaid diagrams

Two quieter changes round out the release. You can now prefix a pattern with ! inside files.exclude or search.exclude to keep specific files visible inside a folder that's otherwise hidden — handy for surfacing something like node_modules/@types without exposing all of node_modules in the Explorer, Search, and Quick Open. Exceptions never hide files outright, and a pattern that matches at any depth can slow searches down, so Microsoft recommends scoping them carefully. Separately, Mermaid diagram rendering in Markdown preview no longer lets edges cross subgraph titles — a title now shifts aside when a line would cross it — and subgraphs using the default theme render outlined rather than filled, which makes nested subgraphs easier to tell apart at a glance.

How to update to VS Code 1.145

If you already have VS Code installed, open the Command Palette and run Check for Updates, or simply restart the editor — it checks automatically on launch. Fresh installs and platform-specific builds (Windows x64/Arm64, macOS Universal/Intel/Apple silicon, and Linux .deb/.rpm/Snap/tarball packages) are available from the official 1.145 release notes page. Developers who want to try remote agent browsing, mixed harnesses, or inline attachments before they reach Stable can switch to the Insiders build, which tracks nightly changes ahead of each monthly Stable release.

Teams running the GitHub Pull Requests extension also get an update worth knowing about alongside the core editor release: its 0.176.0 changelog adds more timeline events (label, milestone, and title changes; review requests and dismissals; force pushes and merge queue updates), issue-variable support in PR description templates, and a prompt to disable auto-merge across a stacked pull request chain.

What's next for VS Code

None of 1.145's headline features ship as finished, stable experiences — remote agent browsing, remote canvases, and mixed harnesses are all Preview, while verbosity modes and inline attachments are Experimental. That pattern matches how Microsoft has rolled out agent-related features across the last several releases: land something behind a setting, gather issue reports for a cycle or two, then flip the default. Given that cadence, expect VS Code 1.146 (currently the Insiders build) to either graduate one of these previews toward Stable or extend mixed harnesses to more entry points in the UI. In the meantime, there are no new deprecations queued beyond the three Local harness tools removed in this release, so teams still on the Local harness have a bit of runway — but Microsoft's own guidance keeps pointing in one direction: toward the Copilot harness as the editor's long-term default.

Written byFelix Moreau

Felix Moreau writes Pandromeda's software coverage and how-to guides. He covers Windows, macOS and Linux updates, the apps people rely on, emulators and developer tools, and turns official documentation into clear, numbered steps that work on the current version.

More from Felix MoreauAll Software & Guides coverage