GitLab Orbit is now in public beta. Context for AI agents in your pipeline.

That sounds abstract, so let me put it more concretely.

An AI agent is good at writing code, but often knows little about the bigger picture. It sees the file it’s modifying, but not how it fits into the rest of your system. In a large monorepo, such an agent gets bogged down: too many iterations, too many tokens, and ultimately work that looks correct but has to be rolled back. Orbit provides the agent with the context it lacks.

Why AI agents get stuck in traditional Git workflows.

Git is designed to keep up with the pace at which people work. You clone a repository, create a branch, and merge your work. That scales perfectly for hundreds of developers worldwide. But it breaks down as soon as each team member runs hundreds of agents at the same time.

GitLab identifies three bottlenecks. The clone tax: an agent clones the entire repository just to read a single file, for every task and every retry. Concurrency collapse: thousands of sessions hit a backend designed for human-scale traffic simultaneously, resulting in unpredictable availability. And no isolation: agents share accounts and a single branch space, causing them to pollute the repo without you being able to trace which agent did what.

In practice, this means increased token usage, slower execution, and higher costs per agent task. If you run more agents simultaneously, that problem escalates rapidly.

What Orbit actually does.

Orbit brings together code, work items, pipelines, deployments, and production signals into a single live graph covering the entire software lifecycle. This allows agents to make decisions based on first-party context rather than isolated, fragmented signals. Engineers query that same graph via the Data Explorer to track changes and investigate incidents. A single source of truth for humans and machines.

Figures from GitLab’s internal tests: agents using Orbit responded up to 11 times faster, were up to 4.5 times more cost-efficient, and produced up to 45 times fewer hallucinations. These are not marginal improvements. Fewer hallucinations mean more reliable code in your deployment pipeline, and lower costs per agent task really add up at scale.

There is also external validation. The client, Compare the Market, tested Orbit A/B against a RAG approach and a context-free variant, using identical prompts and models. Graph-grounded agents placed inline review comments in the correct location in 70 percent of cases, compared to 58 percent for RAG. When tested on 79 real merge requests, the graph approach thus outperformed traditional retrieval.

Orbit also operates as a standalone data product with open APIs, making the same context layer available to third-party agents and external tools.

Governance for agents en next-gen SCM.

In addition to Orbit, GitLab announced two other things that are relevant if you want to stay on top of things:

• Governance for agents (private beta) provides identity, policy, audit, and approval capabilities for every agent action. DevSecOps teams gain real-time visibility into inputs, reasoning, tool calls, and anomalous behavior across the entire organization. If something unexpected happens, the full context is already available: what changed, which policy allowed it, and who signed off on it. For organizations working with requirements related to ISO 27001 or NEN 7510, this is exactly the kind of traceability an auditor wants to see.

• In addition, there is next-generation source code management (private beta): a Git engine rebuilt for agent scale. Early internal results indicate up to 2 times fewer tokens, up to 50 times faster throughput, and up to 1,000 times less network traffic. It runs on the Git protocol, so backward compatibility and auditability are maintained.

The orchestration system that brings all of this together in the daily workflow—the GitLab Duo Agent Platform—has been generally available since January. The number of weekly active users has increased tenfold since then.

What this means for your GitLab strategy.

The questions addressed by Orbit and governance for agents are coming up more and more often: how do you maintain control over AI agents in your pipeline, and how do you ensure that speed doesn’t come at the expense of reliability? Orbit is now available to try in public beta. The other features are still in private beta or are only available through GitLab Ultimate. Want to know how one of these developments fits into your current pipeline setup, or how governance for agents helps with your compliance requirements? Check out what we’re doing with GitLab as a DevSecOps platform or get in touch. Then we’ll work together to figure out what the next step is.