By ByteBot
GitHub Copilot for JetBrains picked up two significant features on August 11: persistent memory that follows you across sessions and repositories, and Ollama as a local model provider. JetBrains users have been waiting on both — VS Code got Ollama BYOK in June, and Copilot memory has been rolling out across surfaces for months. Now it’s here. The setup takes minutes.
Copilot Memory: Stop Re-Explaining Your Stack
If you’ve ever opened a new Copilot chat and typed something like “we use Kotlin, Spring Boot, Postgres, and we never use field injection” — that’s the problem memory solves. Starting now in JetBrains, Copilot can store facts about your repository and preferences about how you work, then recall them automatically in future sessions.
There are two categories of stored information. Repository-level facts cover things like coding conventions, architectural decisions, and build commands. Copilot stores these with citations pointing to the actual code that supports each fact, then validates them against your current branch before using them — so stale facts don’t silently mislead you. User-level preferences are personal: how you want explanations structured, which patterns you prefer, what style of response you find useful. These follow you across every repository and every Copilot agent, without affecting other users in the same codebase.
Retention is automatic: any stored fact or preference unused for 28 days is deleted. Use it, and the timer resets. Repository owners can review and manually remove stored facts through the GitHub settings portal. Enterprise admins can control the feature at the organization level.
To enable in JetBrains: open the Copilot settings panel in your IDE, find the Copilot Memory toggle, and switch it on. Available for Copilot Pro, Pro+, Business, and Enterprise plans.
Ollama BYOK: Your GPU, Your Models, No Cloud Required
The Ollama integration is the more technically interesting addition. You can now connect a local Ollama instance to GitHub Copilot inside JetBrains — and once configured, chat and agent features work entirely offline.
Setup is straightforward:
- Go to Settings → Tools → GitHub Copilot → Model Providers
- Add a new provider and select Ollama
- Enter your localhost endpoint (
http://localhost:11434) — no API key required for a local server - Save. Copilot discovers every model you have pulled locally and adds them to the model picker alongside GitHub-hosted models
For model selection, the guidance is concrete. Qwen2.5-Coder-32B gives the strongest performance for agentic tasks and tool calling, but requires roughly 24 GB of VRAM. The 14B variant at Q4_K_M quantization fits into about 8.7 GB and is the practical choice for a 12 GB GPU card — the sweet spot for quality per gigabyte.
The use case is serious: developers who cannot send employer code to external APIs now have a path that doesn’t require leaving JetBrains or stitching together a custom setup. Air-gapped environments, compliance-sensitive codebases, and cost-conscious teams with capable local hardware all benefit. This was partially available in JetBrains since July and is now complete.
Also Shipping: CLI Auto-Install in the Integrated Terminal
The third item in the August 11 release removes real friction. Copilot CLI — required for terminal-based agent workflows — now auto-installs when you open an integrated terminal in any JetBrains IDE on macOS, Linux, or Windows. Previously, JetBrains users had to install the CLI separately before commands like gh copilot explain would work from inside the IDE. That manual step is gone.
JetBrains Is Catching Up, and That’s Worth Noting
The honest context: JetBrains users have consistently received Copilot features weeks to months after VS Code. Copilot CLI got local model support in April. VS Code got Ollama BYOK in June. JetBrains BYOK expanded in July. Full Ollama support and memory arrive now, in August.
GitHub’s first-party commitment to JetBrains is real — these are not community ports — but the timeline lag has been noticeable enough that some developers moved to alternatives. The August 11 update closes two of the most-requested gaps. If you stayed on JetBrains and kept Copilot, the wait is over.
Enable Copilot memory in your IDE settings today. If you have a capable GPU and privacy concerns about your codebase, the Ollama setup takes less time than this post took to write.
