AI Data Usage and Privacy Controls

7 posts · Mar 2021 – Sep 2026

What this trail covers

GitLab Duo Agent Platform enhances DataOps by automating dbt model generation, including source and workspace models, complete with configurations, tests, and documentation. This significantly reduces manual effort and time for data engineers. The platform is exploring direct integrations with data platforms like Snowflake via Model Context Protocol (MCP) to further automate data structure understanding and model generation. Quality validation is integrated through CI pipelines and AI-generated test queries.

All GitLab trails

Sep 2026Bring your own model to GitLab Duo Self-Hosted with Microsoft FoundryThis post details the implementation steps for connecting GitLab Duo Self-Hosted to AI models hosted in Microsoft Foundry. It covers deploying models within Foundry, recording endpoint URLs, deployment names, and API keys, and installing the AI Gateway. It also provides guidance on choosing models based on GitLab's support matrix and Foundry's catalog, emphasizing feature-level configuration and Azure integration benefits.
Aug 2026Run agentic software delivery inside the boundaries you already trustThis post details the extension of AI Gateway for GitLab Duo Agent Platform to run within GitLab Dedicated environments. It highlights that the AI Gateway runs within the Dedicated single-tenant environment, aligning with data residency. It also emphasizes the ability to bring your own models and connect to providers like Amazon Bedrock, ensuring inference stays within the customer's AWS region. This allows agentic workflows to leverage the trusted infrastructure of GitLab Dedicated.
Aug 2026Confidential AI for GitLab Self-HostedIntroduces the integration of Privatemode AI with GitLab Duo Self-Hosted, enabling confidential AI processing via hardware TEEs. Details the architecture involving a self-hosted AI Gateway and Privatemode proxy for end-to-end encryption and remote attestation. Highlights the benefits for regulated organizations by keeping source code encrypted and within a trusted boundary, addressing challenges of self-hosting LLM stacks. Mentions the use of post-quantum-safe cryptography and an OpenAI-compatible API.
May 2026Claude Opus 4.8 on GitLab: Complex agentic work, less disruptionThis post introduces the availability of Claude Opus 4.8 within the GitLab Duo Agent Platform, accessible via Agentic Chat and other agent workflows. It highlights Opus 4.8's enhanced capabilities for long-horizon agentic execution, leading to more precise and efficient multi-step agent work with fewer interventions. The model also shows improved performance in professional tasks beyond coding, such as document drafting and data analysis. A key technical contribution is the support for mid-conversation system prompts, allowing for dynamic updates to system instructions without invalidating the prompt cache, which is beneficial for asynchronous context arrival. The post also reiterates that Opus 4.8 runs on GitLab Credits and directs users to documentation for credit consumption details.
May 2026Atlassian will train on your data: Opt out with GitLabThis post highlights a significant shift in Atlassian's AI data usage policy, moving to an opt-out-by-default model for training AI offerings on customer metadata and in-app content. It contrasts this with GitLab's unwavering commitment to not train on customer data at any tier. The post details the types of data Atlassian will collect, the implications for different subscription tiers, and the governance challenges posed by opt-out-by-default practices. It emphasizes GitLab's principles of unconditional data commitments, transparency, auditability, and separation of customer data from vendor AI training. It also introduces GitLab Duo Agent Platform for self-managed deployments as a solution for customers who need to keep data within their own infrastructure.
Advertisement
Apr 2026GitHub Copilot's policy for AI training: A governance wake-up callThis post details GitLab's commitment to AI data governance, contrasting it with GitHub's policy change. It emphasizes GitLab's stance of not using customer code for AI model training at any tier, and contractually prohibiting AI vendors from using customer inputs or outputs for their own purposes. It highlights the GitLab AI Transparency Center as an auditable source of information on models, data handling, subprocessors, and retention. The post also discusses the implications for regulated industries, focusing on the need for contractual certainty, auditability, and separation from vendor incentives. It introduces the AI Continuity Plan as a direct response to governance requirements.
Apr 2026GitLab 18.11: Budget guardrails for GitLab CreditsIntroduces subscription-level spending caps configurable in the Customers Portal, which pause DAP access when the cap is reached. Also introduces per-user credit limits (flat and custom overrides) configurable via the GitLab GraphQL API, which pause only Duo Agent Platform credit usage for individual users. Includes email notifications for billing account managers when subscription caps are reached and visibility for group owners/instance administrators on blocked users and cap status.
Mar 2026GitLab 18.10: Agentic AI now open to even more teams on GitLabThis post announces the availability of GitLab Credits for Free GitLab.com teams, allowing them to purchase monthly commitments for agentic AI capabilities. It details the pricing model (pay-per-use, flat-rate for code review at 0.25 credits) and the workflows unlocked (Planner Agent, Developer Flow, Code Review Flow, Fix CI/CD Pipeline Flow, Agentic Chat). It also highlights the value proposition of GitLab Premium, which includes promotional credits and advanced features. The post emphasizes the ease of getting started and the benefits of predictable AI costs.
Jan 2026Introducing GitLab CreditsIntroduces GitLab Credits as a virtual currency for usage-based pricing of agentic AI capabilities, starting with GitLab Duo Agent Platform. This model replaces seat-based pricing to allow pooled credits across an organization for both synchronous and asynchronous agentic AI usage. It details how credits are consumed based on LLM requests, the pricing model (on-demand and annual commitments), and promotional included credits for existing Premium/Ultimate customers. It also outlines cost governance features: a sizing calculator, usage visibility dashboards (Customers Portal and in-product), access controls for teams/projects, and automated usage notifications. The post also describes the upgrade path for existing Duo Pro/Enterprise customers, rolling forward seat commitments to GitLab Credits.
Sep 2025How GitLab Duo Agent Platform transforms DataOpsThis post details the use of GitLab Duo Agent Platform to automate the generation of dbt models for the Reddit Ads dataset. It outlines a manual process for extracting table structure metadata from Snowflake's information schema via SQL queries and exporting it as a CSV. The prompt engineering for GitLab Duo is described, including referencing previous merge requests and the dbt style guide. The output includes generated dbt models, YAML configurations, tests, and documentation. A sample of generated code for the time_zone table is provided, along with an AI-generated validation query for comparing row counts between raw and workspace layers.
Sep 2025Why enterprise independence matters more than ever in DevSecOpsThis post announces the general availability of the GitLab Duo Agent Platform and reiterates GitLab's commitment to enterprise independence, transparency, and developer-first principles in the context of AI integration. It details how the Duo Agent Platform provides visibility into agent decision-making, expanded AI model support for vendor independence, and enhanced governance controls. The post highlights the AI Transparency Center for clear documentation on data governance, privacy, and ethical AI principles, emphasizing that customer data is not used for training. It also discusses model flexibility, the AI continuity plan for evaluating and switching models, and deployment flexibility across on-premises, SaaS, and GitLab Dedicated options. Security and compliance are presented as built-in features, contrasting with fragmented platforms. The post also touches on the importance of open-source community engagement and data governance, ensuring users maintain control over their data and AI processing.
Mar 2021GitLab solicits input on its plans to de-identify service usage dataThis post details plans to de-identify service usage data by limiting access to identifiable information and introducing a new system to de-identify users and other personal information from multi-user instances before the information lands in GitLab's analytics environment. This will allow GitLab to roll up more aggregated, de-identified user-level activity at the account level. It also acknowledges that de-identification may not be perfect for single-user namespaces and solicits input on the design and implementation over the next 30 days.