Microsoft shipped Visual Studio Code 1.123 on June 3, 2026, delivering one of the most AI-forward updates in the editor’s history. The release introduces GitHub-backed synchronization for AI chat sessions, a new /chronicle command set for navigating conversation history, a preview research agent capable of multi-step investigation, and expanded model context windows that can ingest entire codebases. Together, these features signal a deliberate shift from Copilot as a pair programmer to Copilot as a persistent, context-aware development partner.

AI Session Sync: Chat That Follows You Everywhere

The headliner is AI session sync. Until now, every Copilot Chat session lived in local storage, locked to a single machine. Switch from a desktop workstation to a laptop, and your conversation history, inline suggestions, and agent state evaporated. The 1.123 update ties Copilot Chat sessions to a user’s GitHub account, allowing seamless roaming across devices.

Session sync is opt-in and encrypted at rest. Once enabled, VS Code serializes active chat threads, including the full message tree, file context, and any agent-generated artifacts, and stores them in a private GitHub repository automatically created under the user’s account. The repository uses a reserved naming scheme (vscode-copilot-sessions-<userid>) and is hidden from public view by default. When VS Code launches on a new machine, authenticated users see a “Restore previous session” prompt. A single click rehydrates the entire conversation, complete with scroll position and open file tabs.

Early feedback on the VS Code subreddit highlights the practical benefits. Developers moving between a home desktop and a work laptop report no longer needing to bookmark chat threads or keep sticky notes of Copilot-generated SQL queries. “It’s like browser tab sync, but for my train of thought,” one user wrote. Others praise the ability to revisit a refactoring session days later without hunting through commit messages.

Microsoft has baked in privacy controls. Users can delete sessions from the command palette or exclude specific workspaces from syncing via a "chat.sync.blacklist" setting. The feature respects existing Copilot data retention policies; nothing is shared with Microsoft beyond what the GitHub backend already processes.

/chronicle: Command-Line Conversation History

Alongside session sync, VS Code 1.123 rethinks how developers navigate past interactions inside the chat panel. The new /chronicle command—typed directly into the Copilot input box—unlocks a chronological timeline of prompts, responses, and agent actions.

/chronicle list shows a scrollable list of recent sessions, each with a timestamp, a one-line summary generated by a small on-device model, and color-coded badges for activity type (code generation, debugging, refactoring, etc.). Selecting a session expands it inline, showing the full back-and-forth. /chronicle search "upload error" performs a semantic search across all saved sessions, returning relevant snippets ranked by relevance. /chronicle compare lets users diff a current code suggestion against one generated earlier for the same function, making A/B testing of prompt strategies trivial.

The command set supports keyboard shortcuts: Ctrl+Shift+H opens the history panel, and Ctrl+Shift+J jumps to the most recent code suggestion. Power users can chain commands, such as /chronicle list --last 5 --type debugging, to quickly retrieve the context from a late-night bug hunt.

Microsoft says /chronicle is designed to make AI interactions more deliberate and less ephemeral. In internal tests, developers who used the feature reduced redundant prompting by nearly 15%, because they could reference previous solutions instead of rephrasing the same request.

Research Agent: Copilot Learns to Browse and Synthesize

The research agent preview moves Copilot beyond single-turn Q&A into autonomous multi-step investigation. Activated via @research, the agent can accept a high-level query—like “Find the three most common causes of null pointer exceptions in our authentication service and propose fixes”—and spawn parallel sub-tasks that may involve browsing internal documentation, reading project wiki pages, scanning GitHub issues, and even running grep across the repository.

The agent orchestrates other Copilot capabilities: it uses the workspace model to understand the repository structure, invokes the code search index for symbol lookups, and can launch a headless browser session to read external API docs if the user grants permission. Results are presented as a structured Markdown report with clickable citations that jump to the relevant file or web page.

A key constraint is that the research agent does not execute code by default. Users must explicitly approve any command execution or external network request. A status bar indicator shows the agent’s progress—"Researching: 3/5 tasks done"—and a pause button halts activity without losing intermediate results.

Community reaction has been a mix of awe and concern. Developers testing nightly builds praise the agent’s ability to surface obscure configuration errors buried in confluence pages. However, some worry about the cost implications. Each research session can consume significant Copilot quota, especially when browsing external URLs. Microsoft’s pricing page has been updated to reflect tiered limits, with 10 research agent runs per month included in the free tier and 200 in the Pro subscription.

Expanded Context Windows: Feeding Entire Repos to the Model

Underpinning these features is a major backend upgrade. Copilot now supports context windows of up to 2 million tokens on supported models (GPT-5 and Claude 4), allowing it to process roughly the equivalent of 750,000 lines of code in a single prompt. For a typical medium-sized project, that means the model can hold the entire codebase in its attention span without chunking or manual file selection.

VS Code achieves this by leveraging a new indexing service that builds compact representations of all tracked files. On first use, Copilot scans the repository and creates an index in the background. When the user asks a question, the extension retrieves the most relevant segments while keeping the full index as a fallback. This hybrid approach balances response latency and completeness.

In practice, developers report that code completions become startlingly accurate across files. Copilot can now generate a unit test that mocks a dependency defined three directories away without explicit import references. One tester noted, “It felt like the IDE understood the whole architecture, not just the open tab.”

Microsoft warns that large context windows increase GPU compute significantly, so the feature is limited to Copilot Enterprise and paid individual plans during peak hours. Free-tier users get a 128K-token context window.

Additional Improvements and Fixes

Beyond the AI fireworks, VS Code 1.123 polishes the core experience with performance and security enhancements.

Extension Security Overhaul: Following high-profile supply-chain attacks on IDE extensions earlier in the year, Microsoft has tightened the extension installation process. Extensions with native components now undergo a stricter review pipeline, and the marketplace displays a new “Code Signing” badge for packages that pass an extended verification. A runtime sandbox, enabled by default on Windows 11 2026 Update and macOS 15.6, isolates extension processes with reduced privilege tokens, preventing accidental writes to sensitive directories.

Trusted Workspace Improvements: The trusted workspace feature now distinguishes between "trusted," "restricted," and "isolated" modes. Restricted mode disables all extension execution, while isolated mode allows only extensions that have been individually approved. A new UI panel makes toggling modes and reviewing permissions straightforward.

Merge Editor Upgrades: The 3-way merge editor receives a resizeable conflict summary pane and a “resolve all incoming” button. Cherry-pick and revert workflows now surface affected branches in the git history view.

Terminal Link Enhancements: Hyperlinks in the integrated terminal are now clickable by default and underlined with a subtle dotted style. Custom link scanners can be registered via extension API, enabling plugins that detect issue tracker IDs or internal service URLs.

Multi-File Diff View: A long-requested feature lands: users can now select multiple files in the source control view and see a combined diff view. This works natively for git repositories and through the SCM API for other providers.

What the Community Is Saying

Reactions on the VS Code subreddit and Hacker News threads reflect both enthusiasm and pragmatic caution.

Enthusiasts celebrate session sync as a “missing puzzle piece” that makes Copilot feel like a native part of the development lifecycle rather than an ephemeral assistant. Several users shared workflows where they start a complex refactor on a desktop, continue on a tablet via GitHub Codespaces, and review the full session log later on a phone using the VS Code extension for mobile (currently in beta).

Skeptics raise two primary concerns. First, privacy: while Microsoft promises encryption, the idea of AI chat logs stored on GitHub—even in a private repo—gives some enterprise customers pause. A Microsoft representative clarified in a GitHub discussion that organizations can disable session sync globally via a VS Code policy, and that session data never leaves the selected region.

Second, cost: the research agent and large context windows, while powerful, can rapidly burn through quota. Developers on the free tier worry that they will accidentally trigger a hefty compute drain. To address this, VS Code now shows a persistent token count in the status bar when a premium model is active, and a warning dialog pops up when a query would exceed 50% of the monthly free quota.

Extension security upgrades drew near-universal praise. After a wave of malicious extensions injected crypto stealers into popular developer tools earlier in the year, the community has been clamoring for stronger safeguards. The sandboxing and code-signing badges are seen as overdue but effective.

Analysis: Copilot Becomes a Platform

With version 1.123, VS Code crosses a conceptual threshold. Copilot is no longer just a smart autocomplete engine or a reactive chatbot. It is evolving into a persistent, omnipresent layer that remembers your work, learns from your codebase, and acts on your behalf across time and machines.

Session sync and /chronicle turn ephemeral conversations into durable knowledge. The research agent demonstrates that Copilot can orchestrate complex, multi-step tasks. Expanded context windows dissolve the boundary between "the code I showed the model" and "the code we actually have." Taken together, these features position VS Code as the central node in an AI-augmented development workflow.

Competitors are not standing still. JetBrains’ full-line completion and Cursor’s index-based whole-repo understanding have pushed Microsoft to accelerate its roadmap. But VS Code has an unmatchable advantage: its massive extension ecosystem and its deep integration with GitHub. By tying session sync to GitHub accounts, Microsoft turns Copilot into a sticky service that gets more valuable the more repositories you work on.

For Windows developers, the update coincides with the Windows 11 2026 Update that introduced native AI runtime components. The new extension sandbox, in particular, leverages Windows’s AppContainer isolation mode, giving VS Code a security posture that macOS and Linux builds can’t yet fully replicate.

Getting Started

VS Code 1.123 is rolling out as an automatic update for all users. The features described here are available in the stable channel. Session sync requires a GitHub account linked to VS Code and Copilot; the research agent preview is available to all Copilot Individual and Enterprise subscribers. Detailed release notes, including a full list of API changes for extension authors, can be found at code.visualstudio.com/updates/v1_123.

As AI rapidly reshapes developer tooling, this release makes one thing clear: the goal is not to replace the programmer but to create an environment where every keystroke, every chat, and every investigation contributes to a persistent, ever-improving understanding of both the code and the person writing it.