GitLab Orbit unifies code lifecycle in live graph

WHY IT MATTERS: AI coding agents write code well but fail at navigating the systems around it. In large monorepos they waste tokens crawling files, follow wrong threads, miss dependencies, and hallucinate relationships. Across repositories the context window fills before they find answers, and tasks fail outright. Teams end up spending more time fixing agent output than the agent saved them. GitLab Orbit closes this gap by giving agents and engineers a single live graph of code, merge requests, pipelines, deployments, vulnerabilities, and team ownership, with every relationship kept current as work happens.

WHAT CHANGED: GitLab Orbit is now in public beta as a native queryable graph inside GitLab. Agents reason from first-party GitLab data instead of stitched-together tool calls. At Compare the Market, an Orbit-grounded AI reviewer placed accurate inline comments about 70 percent of the time across 79 real merge requests, compared to 58 percent for retrieval-augmented generation. RAG underperformed every other approach tested, including providing no context at all. When Claude Code connects to Orbit through the Model Context Protocol, it stops crawling files and asks the graph direct questions about where code lives, what depends on it, and which tests and pipelines cover it. On the same tasks with the same model, it ran up to 11 times faster, used up to 4.5 times fewer tokens, and produced up to 45 times fewer hallucinations. Engineers can also query Orbit directly through the Data Explorer to trace changes, investigate incidents, and answer cross-system questions that previously required hours of manual reconstruction across multiple tools.

WHAT TO WATCH: GitLab Duo Agent Platform agents can query Orbit natively to triage pipeline failures across the lifecycle, map vulnerability blast radius to owners in minutes, and answer operational questions without custom SQL or dashboard requests. For example, one graph query can return every in-flight merge request that will hit the same failing job across every project in a group. Security teams can get a full owner-by-owner remediation graph the hour a CVE lands. If you run Claude Code against large repos, test the MCP integration immediately. For platform teams building internal AI code reviewers, the Compare the Market results suggest graph-based context may replace RAG as the default retrieval strategy.