In testing, AI agents with GitLab Orbit providing a connected view of the software lifecycle completed tasks with up to 45x fewer hallucinations and up to 4.5x fewer tokens

All Remote – GitLab Inc., the intelligent orchestration platform for DevSecOps, today announced the general availability of GitLab Orbit, a context graph for the software lifecycle. GitLab Orbit connects code, merge requests, work items, pipelines, and vulnerabilities in a single, continuously updated view for AI agents and engineers.

As organizations extend AI across software delivery, agents working in large codebases often lack the information they need to complete a task. Agents often spend multiple iterations gathering context and still work from partial information, producing inaccurate or incomplete answers. Each wrong answer sends the agent back to search files and repositories again, increasing token consumption and cost. With the context available in GitLab Orbit, agents complete tasks more accurately and in fewer attempts, lowering the cost of each task and making AI spend more predictable as organizations scale.

In GitLab testing, agents running with the GitLab Orbit context graph completed the same agentic tasks with up to 45x fewer hallucinations, leading to fewer retries, and up to 4.5x fewer tokens. Orbit enables engineering teams and their agents to answer questions in a single query that once took hours of research or that they could not answer before, such as the impact of a proposed change or which production systems a vulnerability affects. Orbit draws on data that already resides in GitLab, eliminating the need to build additional integrations. It updates automatically as teams ship code and applies the access permissions organizations have already set.

GitLab Orbit works natively with GitLab Duo Agent Platform, and supports third-party agents, including Claude Code, Codex, Cursor, and OpenCode through the Model Context Protocol (MCP) or the GitLab CLI. Since its introduction, more than 3,500 organizations have used Orbit, and their coding agents have run more than 280,000 queries.

Pricing and Availability

GitLab offers GitLab Orbit in query packs that all users and AI agents in an organization share. GitLab account teams size the pack to each organization’s needs. Orbit is now generally available for Premium and Ultimate customers, starting with GitLab.com today and reaching Self-Managed and GitLab Dedicated over the coming weeks. Customers can request a free 30-day trial through their GitLab account team.

Supporting Quotes

• “What GitLab Orbit gave us was something we’d been chasing for a while: the knowledge graph that backs an AI code reviewer that actually understands our codebase, not just the diff in front of it,” said Ryan Harvey, head of AI engineering, Compare the Market. “We tested it against retrieval-augmented generation and a few other approaches across real merge requests, and the gap was clear: better comment placement, better summaries of what actually changed. RAG, which we’d assumed would be the natural solution, ended up performing worse than no context at all. For us, that result spoke for itself.”
• “As organizations extend AI agents beyond code generation into the rest of software delivery, the quality of the context those agents work from matters as much as the models themselves,” said Arnal Dayaratna, research vice president, software development, IDC. “Because GitLab already captures data across the software lifecycle, it is positioned to give agents a more complete view of how software is built and delivered, which can help organizations improve results and better manage the cost of AI at scale.”
• “GitLab Orbit helps organizations get more value from their AI investments,” said Manav Khurana, chief product and marketing officer at GitLab. “AI agents are only as accurate and efficient as the context they work from. Orbit gives every agent in the governed software factory the full record of how an organization builds and ships software, so teams make decisions they can trust, keep spend predictable, and answer questions they never could before.”