GitHub rolled out a three-part update to Copilot for JetBrains on August 11, adding “Copilot Memory,” native Ollama model support, and Codex session receipts that appear in debug logs. The change lets developers decide what the assistant remembers, where its inference runs, and how its actions are recorded—features that matter to anyone worried about privacy, latency, or auditability in AI-assisted coding.

Why the new controls matter

Copilot for JetBrains brings the same AI-driven code suggestions that developers see in VS Code to the JetBrains suite of IDEs. Until now, the plugin behaved like a black box: GitHub’s servers generated suggestions and discarded conversational context after the session. The new layers break that monolith, letting teams pick, case by case, how much interaction stays in memory, whether to offload inference to a local model, and how to surface the assistant’s activity in logs.

Copilot Memory – What stays in your history

“Copilot Memory” persists snippets of conversation across separate chat windows. Ask the assistant to explain a design pattern in one tab and reference it later in another, and the model will recall the earlier exchange. Enterprise admins can toggle the feature, limit the scope of persisted data, or turn it off entirely for compliance-heavy environments.

Ollama integration – Where the model runs

The update adds built-in support for Ollama, an open-source model server you can run on-premises or in a private cloud. Selecting Ollama as the provider shifts inference away from GitHub’s infrastructure, cutting latency and keeping proprietary code inside your network. The plugin shows a simple selector; the underlying model choice stays invisible to the user.

Codex session receipts – How actions are logged

Every Copilot suggestion now generates a “Codex session receipt” written to the IDE’s debug log. New permission modes let administrators record, mask, or omit these receipts. The receipts act as an execution log, showing exactly which prompts were sent, which model answered, and what code was inserted—useful for post-mortem analysis or regulatory audits.

Enterprise knobs and the four-step checklist

Beyond the three features, the update expands enterprise controls: you can gate plugin availability, block access to the Microsoft Copilot (MCP) server, and fine-tune permission-bypass behavior. GitHub recommends four checks before deploying the update:

  • Memory scope: Verify which conversational data the assistant may retain.
  • Model source: Confirm whether inference will run on Ollama or the default cloud service.
  • Permission policy: Review the new permission modes governing what the assistant can do.
  • Execution receipt: Ensure logs capture the desired level of detail for each session.

The trade-off

The added granularity empowers security-focused teams, but it also creates configuration overhead. Admins now juggle three independent settings, and developers may need to adjust workflows for local model latency or altered suggestion continuity when memory is disabled.

Takeaway: GitHub’s August 11 update turns Copilot for JetBrains from a single-track assistant into a suite of controllable components, letting organizations balance convenience, privacy, and auditability on their own terms.