GitHub Copilot for JetBrains: New Model Controls and Chat Fixes

GitHub shipped a changelog update on October 10, 2026 for GitHub Copilot for JetBrains that gives administrators direct control over default AI models, adds a one-click "Fix" action inside diagnostic menus, lets teams turn off automatic MCP server startup for both Copilot and Claude, and reworks several chat and account screens. The same update also ends support for JetBrains IDE 2025.1, so anyone still on that release now needs to upgrade to IDE 2025.2 or later to keep using the plugin.

The release is narrower than GitHub's recent local-sandboxing or model-deprecation announcements, but it matters most to the specific audience that runs Copilot inside IntelliJ IDEA, Rider, PyCharm, WebStorm, and the other JetBrains IDEs, where the plugin has been catching up to the VS Code extension's agent-mode feature set for most of 2026.

What's new in this update, in one list

Four changes define the October 10 changelog entry, according to GitHub's own post:

• Enterprise-managed default models. Organization admins can now use managed settings to pin any available Copilot agent model as the default for new conversations. End users can still switch models themselves through the in-IDE model picker.
• A "Fix" action in diagnostic menus. When the IDE flags an error or warning, the intention menu now includes a Fix option that opens inline chat and asks Copilot to repair the issue directly — using agent mode when it's available, and falling back to ask mode otherwise.
• An MCP auto-start toggle. A new setting lets developers and admins disable automatic startup of Model Context Protocol servers for both Copilot and Claude, so configured tools don't activate until someone explicitly wants them running.
• Chat and account UX changes. Messages are now grouped by conversation turn with navigation controls between turns, user messages can be copied or edited after the fact, account labels are clearer, a dedicated "Switch Account" action was added, and the sign-in screen got a responsive redesign with a direct link to Copilot plans.

Enterprise-managed default models explained

Until now, the model a JetBrains developer saw at the top of a new Copilot conversation was whatever GitHub set as the global default, or whatever that person had last picked. With this update, GitHub extends its managed-settings system — the same mechanism it uses to enforce organization-wide Copilot policy — to cover default model selection specifically for the JetBrains plugin.

An enterprise or organization admin can now designate any model that is available to Copilot's agent mode as the default for every new conversation started inside the plugin. That setting applies org-wide, but it is not a hard lock: individual developers can still override it for a given conversation using the normal model picker in the chat panel. In practice, this is aimed at large engineering organizations that want consistency — for example, standardizing on a specific model for cost-control or compliance reasons while still giving developers an escape hatch when a task genuinely needs a different model.

This sits alongside GitHub's broader model lineup work. Just three days before this changelog post, GitHub added a new Gemini model option to Copilot, and the platform has been rotating through several frontier models from multiple vendors throughout 2026. Managed default-model controls give administrators a way to keep that growing menu from becoming unpredictable across hundreds of JetBrains seats.

The new "Fix" action for diagnostics

The second headline change is aimed squarely at day-to-day coding friction. JetBrains IDEs surface a small intention-action menu (the yellow or red lightbulb) whenever the IDE detects an error, warning, or code-style violation. Previously, acting on a Copilot suggestion for that diagnostic meant switching context — opening chat, describing the problem, and often re-pasting the surrounding code.

GitHub's update adds a "Fix" entry directly to that intention menu. Selecting it opens inline chat with the diagnostic already framed as the task, and asks Copilot to repair the issue. According to the changelog, Copilot will use agent mode for the fix when agent mode is available in that context, and fall back to standard ask mode when it isn't. This is a small interaction change, but it collapses what used to be a multi-step detour — read the warning, open chat, explain it, wait for a reply — into a single click from the exact location where the problem was flagged.

MCP auto-start control for Copilot and Claude

Model Context Protocol (MCP) servers let Copilot's agent mode reach outside the editor — to query a database, call an internal API, or read from a project management tool — through a standardized server process that the IDE launches on your behalf. Until this release, any MCP servers configured for a project would start automatically whenever the relevant session began.

The new setting gives developers and administrators the option to turn that automatic startup off, for both the native Copilot integration and the Claude agent provider inside the same plugin. With it disabled, configured MCP servers remain dormant until someone deliberately starts them. GitHub frames this as a reliability and predictability change: it stops configured tools from silently activating — and consuming resources, hitting external services, or opening network connections — in sessions where nobody actually needs them. For regulated or security-conscious engineering teams, that also doubles as a lightweight control over exactly when an agent can reach external systems at all.

Chat, account, and sign-in changes

Most of the remaining changes are interface refinements rather than new capabilities, but GitHub's changelog lists several that touch the plugin daily:

• Turn-grouped conversations. Chat messages are now grouped by conversation turn, with controls to navigate between turns — useful in long agent sessions where a single task can generate dozens of intermediate messages.
• Copy and edit for your own messages. User messages in chat can now be copied or edited after sending, instead of only being readable as sent.
• Clearer account state. Signed-in accounts get clearer labels and a simpler status display, long account lists are easier to navigate, and a new "Switch Account" action was added for people who work across multiple GitHub accounts or organizations.
• Responsive sign-in page. The sign-in screen was redesigned to work better at different window sizes, with clearer headings, clearer provider choices, a direct link to Copilot plans, and the ability to cancel an in-progress browser authorization.
• Chat discovery nudge. New users now get a one-time reminder pointing them toward the Chat panel.
• Clearer agent placeholder text that highlights what kinds of input — questions, edits, context, or commands — the agent box expects.
• A smoother migration flow. After accepting a migration prompt, selecting "Go Back" now opens a Copilot home session instead of losing context, even if you keep working in the local conversation.

GitHub also lists a set of quality and reliability fixes bundled into the same release: more stable language-server startup, steadier model and provider switching, fewer MCP configuration glitches, better customization refreshes, more reliable file-change handling, and fixes to worktree workflows, along with display fixes in model management, tool-call details, and working-set views.

JetBrains IDE 2025.1 support has ended

Alongside the feature work, GitHub confirmed that the Copilot plugin has ended support for JetBrains IDE 2025.1. The changelog states plainly that you now need JetBrains IDE 2025.2 or later to use the plugin going forward. Developers still on 2025.1 will need to update their JetBrains IDE itself — IntelliJ IDEA, Rider, PyCharm, WebStorm, GoLand, and the rest of the JetBrains family all ship on the same yearly versioning scheme — before installing newer Copilot plugin builds that assume 2025.2 as the floor.

This kind of platform-support deprecation is routine for IDE plugins that track fast-moving IDE platform APIs, but it's worth flagging for any team that pins its JetBrains IDE version for stability: a frozen IDE version eventually means a frozen, aging Copilot plugin version too, missing out on agent-mode improvements that have shipped over the past year.

Why this update matters beyond JetBrains

None of these four changes are dramatic on their own, but together they show GitHub treating the JetBrains plugin as a first-class surface for enterprise AI governance, not just a feature-parity afterthought next to the VS Code extension. Default-model enforcement and the MCP auto-start toggle are both governance levers — they give IT and platform teams more say over how agentic AI behaves inside a JetBrains IDE without removing developer choice entirely.

That matters because JetBrains IDEs are the default for large swaths of enterprise Java, Kotlin, .NET, and Python teams, exactly the kind of organizations that care about model governance, data-boundary controls, and predictable AI agent behavior. It also lands in a busy week for Copilot model and governance news: GitHub pushed a new Gemini 4 Argon model option and broader enterprise managed-settings enforcement updates within days of this JetBrains changelog post, suggesting a coordinated push on administrative controls across the whole Copilot product line, not just one IDE integration.

For teams still deciding how Copilot for JetBrains stacks up against alternatives, this update is also a useful data point in the ongoing comparison between GitHub Copilot and Microsoft Copilot and in cost conversations like Claude Code versus GitHub Copilot pricing — the more granular GitHub's admin controls get, the more its pitch to large engineering organizations starts to look like the identity and governance layer vendors sell alongside the coding assistant itself, not just the assistant.

Who should pay attention to this release

Three groups get the most direct benefit from this specific changelog entry:

• Platform and DevEx teams running Copilot at enterprise scale, who now have a cleaner way to standardize the default agent model across every JetBrains seat without disabling developer choice.
• Security and compliance-minded teams that want to stop MCP servers — which can reach external APIs, databases, or other tools — from auto-starting in every agent session by default.
• Individual developers who will feel the smaller day-to-day improvements most: the one-click Fix action on diagnostics, the ability to edit a chat message instead of retyping it, and turn-grouped conversations that make long agent sessions easier to scan.

How to update and what to check next

There's no separate opt-in required for most of these changes — they ship with the plugin update itself. To get them:

1. Confirm your JetBrains IDE is on version 2025.2 or later. If it isn't, update the IDE first; the plugin's new baseline depends on it.
2. Update the GitHub Copilot plugin from inside your JetBrains IDE's plugin marketplace, or check the current release directly on the JetBrains Marketplace versions page for GitHub Copilot.
3. If you administer Copilot for an organization or enterprise, look for the new default-model managed setting in your Copilot enterprise policy configuration, and decide whether to pin a default model or leave the existing behavior in place.
4. If your team relies on MCP-connected tools, decide whether automatic startup is still the right default, or whether the new manual-start toggle better fits your security posture.
5. Report bugs or friction with the new release through the plugin's own feedback channel, the copilot-intellij-feedback issue tracker on GitHub, rather than general GitHub support.

GitHub publishes these JetBrains-specific changelog entries on a rolling basis — a similar round of JetBrains plugin improvements landed in February 2026, and agent mode itself reached general availability for JetBrains, Eclipse, and Xcode in mid-2025 — so teams that standardize on JetBrains IDEs should expect this cadence of incremental admin and UX updates to continue rather than treat October's release as a one-off. GitHub's own documentation hub, docs.github.com/en/copilot, remains the canonical reference for configuring managed settings, MCP, and agent-mode behavior as these controls evolve.

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