GitLab Orbit GA: a context graph for AI coding agents

GitLab made Orbit generally available on October 29. Orbit is a context graph for AI coding agents: one live view of your code, merge requests, work items, pipelines and vulnerabilities that any agent can query. It is on GitLab.com now for Premium and Ultimate. For a small team that runs Claude Code or Cursor against a GitLab repo, the question is simple. Is better context worth a new meter?

What actually happened

Per GitLab's launch post, Orbit builds a graph from data that is already in GitLab. No extra integrations. It updates as you ship, and it applies the permissions your groups and roles already have. Orbit entered beta in June. GitLab says more than 3,500 organizations used it and their coding agents ran more than 280,000 queries before GA.

Agents connect three ways. GitLab Duo Agent Platform queries it natively. Claude Code, Codex, Cursor, OpenCode, Gemini CLI and other MCP-compatible agents connect through the Model Context Protocol or the GitLab CLI (glab orbit setup). Custom agents use a REST API.

Self-Managed and GitLab Dedicated get Orbit "over the coming weeks." No dates.

Pricing is a shared pool of queries. A query is one request to the graph, from a person, an agent or the API. GitLab lists a pack at $30 in GitLab Credits per user per month, which adds 1,000 queries to the pool. Overage is about $0.03 per query. Existing customers can ask their account team for a free 30-day trial.

The headline numbers are GitLab's own. The blog says agents on Orbit needed "up to 45x fewer retries" and up to 4.5x fewer tokens on the same tasks with the same model. The press release calls the first number "hallucinations." Neither names the task set or the model. The more useful data point comes from a customer: Compare the Market tested four context methods on 79 real merge requests. Orbit matched about 70% of the expected review comments. RAG matched about 58%, which GitLab says was worse than no added context at all.

Why it matters for your business

Context is now the product, and it has a price. Your agent bill used to be tokens. With Orbit it is tokens plus graph queries. An agent loop that checks the graph on every step can burn 1,000 queries fast. At three cents each, a busy week is real money. Log query counts per task in the trial before you buy packs.

Your homegrown RAG may be the weak link. The Compare the Market result is one test, but it matches what we see. Chunked code search pulls the wrong file with confidence. If your review agent runs on a vector index of your repo, compare it against plain context and against Orbit on 20 of your own merge requests.

Keep the agent swappable. Orbit speaks MCP, so the graph stays the same when you change agents. That is the right shape. Wire it as one MCP server among several, not as a reason to standardize on one coding tool.

Sources: GitLab via Business Wire, GitLab blog: Orbit is generally available.