JetBrains版GitHub Copilotに永続メモリとOllamaローカルモデルが追加Copilot memory and Ollama in GitHub Copilot for JetBrains

GitHub has updated Copilot for JetBrains with a set of changes that bring the plugin closer to feature parity with the Visual Studio Code experience, most notably persistent memory, the ability to run local models through Ollama, and additional enterprise controls. The update matters for developers who work primarily inside JetBrains IDEs such as IntelliJ IDEA, PyCharm, WebStorm, GoLand, and Rider, because Copilot's newer capabilities have often landed on other editors first. According to the changelog, the release also improves everyday chat workflows and resolves reliability issues across Model Context Protocol (MCP) servers.

The headline addition is persistent memory, sometimes referred to as Copilot memory. In practice, this appears to let the assistant retain context and preferences across sessions rather than treating each conversation as a blank slate. For repeated tasks, coding conventions, or project-specific details, persistent memory is intended to reduce how often a developer must restate the same information. Memory features of this kind typically raise questions about what is stored, where, and for how long, which is likely one reason the update pairs the capability with expanded administrative controls for organizations.

The second major change is local model access via Ollama. Ollama is an open-source tool that runs large language models directly on a developer's own machine, packaging models such as Llama, Mistral, and other open-weight options behind a simple local server. By connecting Copilot for JetBrains to an Ollama endpoint, users can route certain requests to a model running on their own hardware instead of a hosted service. This is significant for teams with strict data handling requirements, offline or air-gapped scenarios, or a preference to experiment with open models. It is worth noting that local models generally depend on available compute, and smaller models running on a laptop will not necessarily match the quality of large cloud-hosted systems. The changelog frames this as added flexibility rather than a replacement for GitHub's default models.

Enterprise controls round out the core additions. GitHub has steadily built out administrative tooling for Copilot, including policy settings, content exclusion, and usage governance, and this update extends more of that management surface to the JetBrains plugin. While the changelog excerpt does not enumerate every new control, the emphasis on enterprise governance aligns with how organizations increasingly evaluate AI coding tools: not only on output quality but on the ability to constrain where data flows and which capabilities are enabled for which users. The combination of memory, local models, and stronger controls suggests GitHub is trying to make the JetBrains client viable for regulated and security-conscious environments.

The update also touches the Model Context Protocol layer. MCP is an open standard, originally introduced by Anthropic and since adopted across a range of tools, that defines how assistants connect to external data sources, services, and tools through dedicated servers. Copilot's support for MCP allows the assistant to reach beyond the code in the editor to pull in additional context or trigger actions. Reliability fixes across MCP servers should, in principle, make those integrations more dependable, reducing failed connections or inconsistent behavior when multiple servers are configured. For developers who have wired Copilot into internal systems through MCP, stability improvements are often as valuable as new features.

Alongside these larger items, the release includes general improvements to everyday chat workflows. The changelog does not fully detail these refinements, but such updates commonly address the friction points that accumulate in daily use, such as smoother conversation handling, clearer responses, or fewer interruptions during a coding session.

For context, this release fits a broader industry pattern in which AI coding assistants are converging on a similar feature set: cross-session memory, support for both hosted and local models, standardized tool integration through protocols like MCP, and deeper enterprise administration. Competing tools and IDE integrations have pursued comparable directions, and the ability to swap between cloud and local models via runners like Ollama has become a recurring theme. JetBrains itself has been developing its own AI features, so Copilot's continued investment in the JetBrains ecosystem reflects an environment where multiple assistants coexist within the same IDEs.

Developers using Copilot for JetBrains should be able to access these changes by updating the plugin to the latest version, with the local model and memory features likely subject to organizational policy where enterprise controls apply. As with any changelog entry, the practical impact will depend on individual configurations, hardware, and how administrators choose to enable the new options.