GitLab makes Orbit generally available to connect code, pipelines and vulnerabilities for agents

GitLab made Orbit generally available to provide project context to agents.

GitLab announced general availability of Orbit on October 29. It describes Orbit as a graph that connects code, merge requests, work items, pipelines and vulnerabilities, with updates as the underlying information changes. The launch starts on GitLab.com for Premium and Ultimate customers. GitLab says Self-Managed and GitLab Dedicated customers are due to receive it over the following weeks. [GitLab announcement](https://about.gitlab.com/press/releases/2026-10-29-gitlab-announces-orbit-general-availability/)

A connected view can answer a different question from code search

A code search can find every place a library is imported. An engineer tracing a vulnerability also needs to know which changes introduced the dependency, what pipeline built it and where the resulting artifact went. The graph provides relationships that an agent could otherwise have to reconstruct from separate queries. The completeness of that answer still depends on the underlying data and which projects the requester is allowed to see.

GitLab says Orbit works with its Duo Agent Platform and can also be used by third-party agents including Claude Code, Codex, Cursor and OpenCode through MCP or the GitLab CLI. Organizations share query packs across users and agents, according to the announcement. An administrator evaluating costs would therefore need to look at graph queries and model usage together. [Availability, integrations and query packs](https://about.gitlab.com/press/releases/2026-10-29-gitlab-announces-orbit-general-availability/)

How to interpret the reported improvements

GitLab reports up to a 45-fold reduction in hallucinations and up to a 4.5-fold reduction in token use in its tests. The phrase “up to” matters because those figures describe the strongest reported differences, not a typical reduction across all workflows. A team can test a representative set of known repository questions, grade the answers against evidence and count graph requests and model tokens per accepted answer.

The launch gives customers a way to test whether exposing software delivery relationships cuts the time an agent spends assembling context. Its value depends on how well the graph reflects the systems that a team actually runs.