The Gemini large language models (LLMs) that are used by Google Cloud products are trained on datasets of publicly available code, Google Cloud-specific material, and other relevant technical information in addition to the datasets used to train the Gemini foundation models. Models are trained so that responses for Google Cloud products are as useful to Google Cloud users as possible.
Additionally, Gemini in the Google Cloud console includes basic user contextual information (project ID, product area, page title, and organization ID) when giving responses. Google Cloud CLI (gcloud CLI) commands and other code may incorporate user context directly into the generated code snippet. Gemini doesn't persist contextual information.
Gemini for Google Cloud doesn't use your prompts or generated responses for training or fine-tuning our underlying models. Any data sent to the Gemini models is used strictly for serving a response to the request, and unless instructed by you, isn't stored. There are no controls available to filter or block certain information from being sent to Gemini for Google Cloud.
Google Cloud products serve Gemini requests as close to the user as possible where capacity is available.
How and when Gemini cites sources for Google Cloud products
Gemini LLMs for Google Cloud products, like some other standalone LLM experiences, are intended to generate original content and not replicate existing content at length. We've designed our systems to limit the chances of this occurring, and we continue to improve how these systems function.
If Google Cloud products using Gemini directly quote at length from a web page, they cite that page. For answers with URLs, Google Cloud products using Gemini lets users see and, in some cases, click to navigate directly to the source page.
When generating code or offering code completion, Google Cloud products using Gemini provide citation information when they directly quote at length from another source, such as existing open source code. In the case of citations to code repositories, the citation might also reference an applicable open source license.
How Google Cloud products using Gemini help protect you with generative AI indemnification
Generative AI Indemnified Service covers certain Google Cloud products that use Gemini.
For these products, if you are challenged on copyright grounds after using content generated by Gemini, we assume certain responsibility for the potential legal risks involved.
For full details about the indemnity, see our Service Specific Terms, or read our blog post on this issue.
Gemini Code Assist
To allow for better code generation in IDEs, Gemini Code Assist gathers contextual information from the file that you're actively using in your IDE as well as other open and relevant local files in your project.
When working with Gemini Code Assist in your IDE, Gemini lists your project files (the context sources) that were used as reference to generate responses to your prompts. Context sources are shown every time you use Gemini chat.
You can prevent Gemini Code Assist from suggesting code that matches cited sources by adjusting settings in your IDE (VS Code, Cloud Shell Editor, and Cloud Workstations).
Code customization lets you get code suggestions based on your organization's private codebase directly from Gemini Code Assist. To learn more about code customization, and how we provide security when accessing and storing your private code, see the Gemini Code Assist overview. To configure and use code customization, see Configure and use Gemini Code Assist code customization.
For more information about Gemini Code Assist security controls, see Security, privacy, and compliance for Gemini Code Assist on Google Cloud.
For information on where Gemini Code Assist serves from, see Gemini serving locations.