Des rames de métro sans conducteur mises à lessai à Shanghai

Different LLMs have different strengths, so setting up your AI architecture with multiple models for specific use cases can be a path to success. However, it’s important to ensure that DevOps providers are transparent about the LLMs they utilize for their AI features as well as details about where the LLMs are hosted.

GitLab Duo features aren’t powered by a single model. We have built GitLab Duo with the flexibility to use the model that provides the right result for each use case. We continue our commitment to transparency by clearly identifying the models powering GitLab Duo features in our publicly available documentation.

GitLab Duo is powered by third-party models hosted on cloud infrastructure, and the vendors of these models and the terms on which they provide services to GitLab were chosen as they support GitLab’s commitment to privacy and the protection of customer intellectual property.

We list all our subprocessors clearly on our subprocessors page, and customers can sign up to be notified when updates are made to this page.

GitLab does not currently train its own models. No inputs to or outputs from GitLab Duo features are currently used to train any GitLab models. Our AI vendors are under data protection agreements with GitLab that prohibit the use of prompts or output associated with GitLab's customer ID for their own purposes, except to perform their independent legal obligations.

With GitLab Duo, your inputs remain your content. GitLab makes no claim of ownership in your input. See our AI Functionality terms for more details.

While the legal and regulatory landscape related to AI-generated output is developing, GitLab’s position is clear. GitLab does not claim ownership of any output generated by GitLab Duo. Output generated by GitLab Duo can be used at your discretion. See our AI Functionality terms for more details.

GitLab will indemnify GitLab Duo customers and protect their right to use output generated by GitLab Duo, including covering defense costs arising from claims that GitLab-Duo-generated output infringes a third party’s intellectual property rights. Similarly, GitLab will also pay amounts of a final adverse judgment or approved settlement related to those claims. GitLab protects these customers’ right to use output generated by GitLab Duo as long as they:

1. have not modified the output;
2. have a valid right to use their inputs;
3. have paid for the AI feature(s); and
4. have no knowledge that the output is infringing before using or otherwise relying on it.

Customers do not need to enable or activate any filters to receive this protection. See our AI Functionality terms for more details.

DevSecOps providers should share specific documentation about how their AI features use your data. Here are the relevant resources for GitLab customers:

• GitLab Subscription Agreement
• AI Functionality Terms
• GitLab Testing Agreement
• GitLab Privacy Statement
• Acceptable Use Policy
• GitLab Duo Documentation

GitLab will continue to retain its Service Level Objective (SLO) and monitor GitLab.com to report actual monthly Service Level Availability (SLA).

GitLab has published AI Ethics Principles for Product Development to help guide GitLab as we build and iterate on our AI features.

GitLab does not publish GitLab Duo’s acceptance rates because measuring effectiveness with AI features is not a straightforward calculation.

The measurement we are interested in evaluating is how using GitLab Duo can contribute to meeting customer business objectives. An example to consider: how would GitLab Duo, as part of your GitLab workflow, help your developers complete feature requests faster?

For more information on our approach, please see this blog post.

GitLab does not currently develop or train its own AI models – please see above for our AI documentation, as well as our AI ethics principles and other resources. GitLab’s current AI features do not qualify as high risk AI systems under the AI Act, and comply with the Act’s requirements to the extent that they constitute limited-risk systems. GitLab provides a number of assurances with respect to our products’ compliance with applicable law, such as GitLab’s Privacy Statement, Subscription Agreement, AI Functionality Terms, and Testing Agreement (which applies to GitLab’s experimental and beta AI features), all available here.

Workflow cannot currently push any changes to production without human intervention.