VS Code Local ends in February 2027: How to move to Copilot

When does Local end, and can I still use it?

Microsoft deprecated the Local harness in VS Code 1.146 Stable, released on November 11, 2026. The release notes give a three-month transition: Local remains usable but is functionally frozen, while new capabilities focus on Copilot. It becomes unavailable after those three months, in February 2027. No exact removal day or time zone is specified, so avoid scheduling migration for the end of that month.

A harness coordinates an agent’s context, tool calls and approvals. Local is the legacy editor-specific implementation; Copilot uses the shared Copilot SDK runtime. The harness, selected model and execution host are separate choices: selecting a Claude model does not select the Claude harness, and selecting Copilot does not automatically move local files to the cloud.

This guide checked the release notes and current migration documentation on November 13, Beijing time. If you already start sessions in Copilot, first confirm the current harness. Copilot has been the default for new sessions since 1.143, but existing conversations and customizations may still need attention.

Compare one familiar task before and after the switch

Update to your organization’s current approved VS Code release, use the GitHub account with your approved Copilot access, and back up customization files. Model and feature availability still depends on account access, plan, organization policy and feature prerequisites. The deprecation announcement does not promise uniform pricing or availability in every region.

Choose a small task that already works in Local. Record its expected result, required tools, instructions and approval flow. After migration, compare actual edits and test results on that task; opening a new window alone does not establish a working setup.

Choose New Chat in the editor’s Chat view, or New in the Agents window and select the project. In Session Target, select Copilot, then choose an available model and agent. For an existing conversation, inspect its current harness rather than inferring it from the model name or window.

Start customization migration with the Migrations list

Run Chat: Open Customizations, select Migrations, and review the detected items. Choosing Copilot does not convert everything automatically. Prioritize workflows you use daily, then remove old copies only after verifying that they are no longer needed.

Convert prompt files to skills and recheck agent, model and tool choices that conversion does not preserve. For personal agents, instructions and skills, review profile storage and search locations. Hooks have no automatic conversion: check their events, script inputs and outputs, and failure behavior manually.

For MCP servers, recheck configuration, authorization, available tools and sandbox behavior. Stored editor secrets and input prompts do not transfer automatically; review configurations using ${input:...} or ${command:...} carefully. Do not replace a placeholder with a literal credential in a shared or personal configuration file for convenience.

Original and migrated files do not synchronize. Keep backups until verification is complete and avoid duplicate active copies. Migrate performs supported file moves and conversions. Migrate with Agent remains Experimental and uses Copilot credits for agent-assisted adaptation; it is not a guaranteed free automatic repair. Extension-provided tools also need a connected editor window where the extension is running.

Continue an existing Local conversation in Copilot

Open the original Local session in the editor and select Copilot in Session Target in its input area. Handoff carries conversation history and context to the destination. Check the available model, tools and permissions before continuing, because they can differ. This is not a two-way switch between Local and Copilot.

When organization policy requires Copilot, opening an unarchived Local conversation can move it automatically. On a host with import support, history and the unsent draft are preserved and the original is archived as read-only. Without import support, Continue in Copilot starts a new conversation with the transcript attached as context, preserves the draft, and leaves the original accessible. Neither path submits a prompt automatically.

If earlier work is missing, check the original workspace, execution host, and Archived or Done filters before treating it as lost. Local sessions are editor-only. A Copilot session can continue through Open in Agents or Open in Editor, but its Agent Host must keep running, and editor-provided tools still need a connected editor.

Separate organization requirements from temporary exceptions

Pilot the projects, integrations and execution environments your team actually uses before expanding deployment. For VS Code 1.143 or later, the official requirement path sets vscode.temporaryAllowLocalHarness and vscode.temporaryDefaultToLocalHarness to false in enterprise-managed settings, not ordinary user or workspace settings.

temporaryAllowLocalHarness: false prohibits the Local, Claude and Codex harnesses and takes precedence over the temporary Local default. To encourage adoption without prohibiting other harnesses, the ChatEditorPreferCopilotHarness device policy can select the default for new editor chats. Administrators should check the intended scope before choosing a control.

Requiring Copilot does not enable sandboxing or add permission-deny rules; deploy those controls separately. Setting vscode.temporaryDefaultToLocalHarness temporarily to true suppresses automatic routing, but does not override temporaryAllowLocalHarness: false or change explicit, remembered or current harness choices.

The temporary default is also being removed in February 2027. Local cannot enforce Copilot runtime-only permission and sandbox rules, so retaining the same managed configuration does not establish equivalent protection. Use fallback only where organization policy permits and an administrator has reviewed that gap, while tracking the blocker and a supported replacement.

Verify the host that actually runs commands

Run the familiar task in the intended Copilot session. Inspect file diffs, command results, tests, and whether the migrated skill, hook or MCP tool actually works. Check shell initialization as well; the agent’s completion summary is not enough.

Use Developer: Policy Diagnostics to inspect effective policies and their sources. Then run /sandbox policy in the Copilot session to check the execution host, sandbox status, and file and network boundaries. Approvals decide whether an action needs confirmation; sandboxing limits what supported processes can access. An approval dialog alone does not establish effective isolation.

For Remote SSH, WSL and Dev Containers, verify prerequisites and paths on the host or container that runs the commands. Local checks do not establish remote readiness. An administrator can arrange one allowed action and one harmless action that policy should block, using a disposable test path, to verify enforcement.

This article reviews public migration documentation; it does not migrate any account’s configuration or validate an execution host. Investigate feature, model and policy restrictions through the relevant account and organization rules. Changing network egress cannot extend the deprecation deadline, grant plan eligibility or replace permission controls.