What happened
GitHub shipped an update to Copilot for JetBrains IDEs on August 11, and the two changes worth knowing are both about where Copilot gets its information from.
The first is Copilot memory. Copilot can now “retain and recall useful information across agent chat sessions,” so you stop re-explaining the same project details or preferences every time you open a new chat. It’s controlled by a Copilot Memory toggle in the Copilot settings portal, meaning you can turn it off. GitHub’s changelog links out to a separate “About Copilot memory” page for more detail, but doesn’t itself say how long that memory persists or whether it’s kept separate between different projects you work on.
The second is Ollama as a BYOK (bring your own key) provider. BYOK means you connect a model you run or pay for yourself instead of only using GitHub’s built-in options. With Ollama, that model can be running locally on your own machine rather than in the cloud. The changelog confirms “provider configuration and model selection throughout the JetBrains experience” but doesn’t spell out the exact setup steps, so check GitHub’s own BYOK provider documentation rather than assuming it works like the other providers you may have already configured.
The same update adds enterprise managed settings, giving organization admins centralized control over which plugins are allowed, which outside tools Copilot can connect to, when it can skip asking for permission, and what usage data gets sent back to GitHub. That part is aimed at IT admins, not individual developers. The update also makes Codex sessions show up in Copilot’s own activity logs and lets Copilot’s command-line tool install itself automatically from JetBrains’ built-in terminal on macOS, Linux, and Windows. Note that this specific rollout is for JetBrains IDEs (IntelliJ, PyCharm, and similar), not VS Code.
Why it matters
Persistent memory solves a real daily annoyance: repeating your stack, conventions, and preferences at the start of every chat. But the changelog doesn’t say whether memory is scoped to one project or shared across every project tied to your Copilot account. That distinction matters if you work across multiple unrelated codebases or client projects from the same JetBrains setup. Without project-level scoping, a preference, naming convention, or detail from one project could surface in a chat about a completely different one, and there’s no way to tell from this changelog alone whether that’s how it behaves.
Ollama BYOK is a real option for anyone who wants a local or open-weight model (one whose underlying files you can download and run yourself, instead of only accessing it through a company’s servers) available inside Copilot’s own interface, instead of switching tools entirely, whether that’s for cost, or for keeping certain code from leaving your machine. Whether it actually keeps that code local depends on how you configure it, and GitHub hasn’t published the specific steps in this changelog.
Who should care
This matters most if you use GitHub Copilot inside a JetBrains IDE and juggle more than one client or codebase from the same account, since memory scoping is the thing to verify before you lean on it. It also matters if you’ve wanted to test a local model through Copilot’s own chat instead of a separate tool. It matters less if you’re a VS Code-only Copilot user, since this specific update doesn’t cover that IDE, or if you only ever work in a single project where cross-project memory leakage isn’t a concern.
What builders should do next
Test memory scoping directly before you trust it with anything sensitive. In one JetBrains project, tell Copilot a specific, made-up preference it wouldn’t otherwise guess, for example, “in this project, always use 4-space indentation and named exports only.” Close that chat. Open a new chat in a different, unrelated project in the same account and ask Copilot to write a small function, without mentioning the preference. Check two things: did the unrelated project’s suggestion stay clean of that preference (correctness), and did the original project’s next new chat still remember it (whether memory is actually working as intended). If the preference leaks into the unrelated project, treat memory as account-wide, not project-scoped, and turn off the Copilot Memory toggle for any account you use across multiple clients.
If you want to try Ollama as a provider, install and confirm Ollama is running locally first, then look for the BYOK provider option in Copilot’s settings inside your JetBrains IDE rather than guessing at a configuration path. Check GitHub’s BYOK documentation for the exact address it expects for your local Ollama installation, and the model-name format, since the changelog itself doesn’t include that detail.
End of article