Copilot for JetBrains adds Ollama BYOK and memory

Two features shipped in GitHub Copilot for JetBrains today, and they pull in opposite directions. Ollama is now a BYOK provider, which means your IDE can drive a model running on your own hardware. And Copilot memory now retains information across agent chat sessions. The first one you should turn on. The second one you should decide about deliberately, before someone else does.

What actually happened

Per the GitHub Changelog, Ollama joins the bring-your-own-key provider list in Copilot for JetBrains, with provider configuration and model selection available throughout the JetBrains experience. Copilot memory "can now retain and recall useful information across agent chat sessions" so you stop re-explaining project structure and preferences every time. It's governed by a Copilot Memory toggle in the Copilot settings portal. The same release added enterprise managed settings, expanded Codex workflows, and automatic Copilot CLI installation in the terminal.

The changelog doesn't specify which JetBrains IDEs, which Ollama models, or where memory is stored. Worth checking before you assume — "local model" and "local memory" are separate claims, and only one of them was made.

Why it matters for your business

Ollama as a BYOK provider is the more durable win. Ollama runs open-weight models on hardware you own, so a JetBrains developer can point agent sessions at a model on a workstation or an internal box. That matters in two specific situations: code you're contractually barred from sending to a third-party inference endpoint, and cost-sensitive bulk work — test scaffolding, refactors, doc generation — where a local 20-something-billion-parameter model is good enough and the marginal token is free. Keep the hosted frontier models for the hard problems. This is the same portability argument we make about every model layer: make it a dial, not a dependency.

Copilot memory needs a different conversation. Cross-session memory is genuinely useful — it's why agents feel competent on day three instead of day one. It's also a new persistent store containing whatever your developers happened to say. Client names. Internal architecture. Credentials pasted into a chat by someone in a hurry. Before you leave that toggle on, answer three questions: where does it live, who can read it, and how do you wipe it when a project or a contractor ends. If your vendor can't answer, the toggle is the answer.

Small-team playbook: Ollama on, memory on per-project with a documented reset, and enterprise managed settings used to make both of those the default rather than a per-developer choice.

Sources: GitHub Changelog — Copilot memory and Ollama in GitHub Copilot for JetBrains, Ollama.