Orbit turns GitLab data and code into one graph. Agents and people query that graph to answer questions that would take many GitLab API calls. The data model page describes what the graph holds.
How data gets in
Orbit only reads from GitLab. It never connects to the GitLab database. Siphon copies the GitLab tables we need into a data lake. The SDLC indexer turns those rows into nodes and edges. The code indexer fetches repositories through GitLab and adds the code graph. A dispatcher schedules work for both indexers.
SDLC indexing and code indexing walk through each indexer. Schema evolution covers schema changes with no downtime. Resiliency covers how indexing recovers from failures.
How a question gets answered
Every request goes through GitLab first. GitLab checks who the caller is and what they can see. Orbit then runs one bounded query inside that scope. GitLab checks each result again before it goes back. Results show the graph as of the last indexing run.
Querying follows a request from start to end. Security explains each check.
Why we built it this way
We wanted deploying Orbit to add no operational load to GitLab. Orbit runs every query on its own store, so the GitLab database never serves graph queries. We built that store on data infrastructure that GitLab already runs.
ADR 000 compares dedicated graph databases, like Neo4j, with the databases GitLab already runs. Reusing one of those meant one less database to deploy and maintain.
Where it runs
Orbit runs on GitLab.com, GitLab Dedicated, and GitLab Self-Managed. It ships separately from GitLab. Orbit on GitLab Self-Managed covers the setup.
More pages
- Orbit CLI brings Orbit to your terminal and to your AI agents.
- Monetization explains how Orbit reports billable usage and checks quota.
- Testing shows how we prove a change works before it ships.
- The ADRs record our main decisions and why we made them.