VS Code 1.143 makes the Copilot harness the default for new sessions
New chat and agent sessions now run on the Copilot SDK harness automatically, with no listed opt-out. The release also adds model comparison and an experimental on-device model.
VS Code 1.143, released on 19 October, makes the Copilot harness the default for every new chat and agent session. The release notes state it plainly: “The Copilot harness is now the default for new sessions in VS Code.” Unlike 1.141’s opt-in sandbox setting and 1.142’s Windows-only harness deprecation, this is the actual switch, and no opt-out setting is listed.
The same release adds a preview “compare agents” mode in the Agents window (sessions.chat.compareAgents.enabled), which runs the same task across multiple models in parallel, optionally scores the attempts with a Judge model, and lets you combine the results into one solution. It also adds an experimental on-device model, MAI Code 1.1 Flash, which needs GitHub sign-in, an NVIDIA GPU with 96GB or more of memory, and a local llama.cpp build, in exchange for responses that don’t consume AI credits.
Why it matters: if your team hadn’t deliberately opted into the Copilot harness, new sessions will use it anyway from this version on. Worth checking your settings before upgrading if you depend on the classic chat behaviour, especially under any enterprise sandbox or permission policy.
The caveat: the on-device model’s 96GB VRAM requirement puts it out of reach for most laptops; it’s a feature for workstations, not a general local-inference option yet.