<!--IssueSummary start--> <details> <summary> Everyone can contribute. [Help move this issue forward](https://handbook.gitlab.com/handbook/marketing/developer-relations/contributor-success/community-contributors-workflows/#contributor-links) while earning points, leveling up and collecting rewards. </summary> - [Label this issue](https://contributors.gitlab.com/manage-issue?action=label&projectId=278964&issueIid=598149) - [Close this issue](https://contributors.gitlab.com/manage-issue?action=close&projectId=278964&issueIid=598149) </details> <!--IssueSummary end--> > [!warning] > > **There will be bugs.** Queries may have issues--timeouts, performance, etc--as the team works through scaling challenges. Please help [report these](https://gitlab.com/groups/gitlab-org/-/work_items/21852). > > **Don't use Orbit for primary key look ups or simple use cases** like 'summarize this issue.' This is for LARGER tasks, research tasks, analytics task, and advanced understanding. GKG indexes the GitLab SDLC as a knowledge graph - code, pipelines, MRs, issues, and vulnerabilities, and the relationships between them. It makes that structured context available to AI agents in real time. The difference from standard Duo Agent or RAG: the agent traverses relationships, not just files. "What calls this function?" "Which files correlate with failing pipelines?" "What breaks if I change this service?" - these are answerable now, on the monolith. ## Get Set Up **1. Orbit Agent in DAP side panel** - Go to User Preferences -\> **Enable Orbit in GitLab Duo** **_please note, Orbit Agent is the only full integration at the current moment, we are working on complete DAP integration_** {width="865" height="257"} {width="328" height="93"} **2. External agents (Claude Code, Cursor, Codex, Duo CLI, etc.)** - one-time `glab` setup: ```bash brew install glab # 1.94+ glab auth login glab orbit remote status # should report 6 components healthy ``` Then add the MCP server to your agent's config (`~/.claude.json`, `~/.codex/config.json`, etc.): ```json { "mcpServers": { "gitlab-orbit": { "command": "npx", "args": ["mcp-remote", "https://gitlab.com/api/v4/orbit/mcp"] } } } ``` Restart the agent. You get `query_graph` and `get_graph_schema`. Full docs: https://docs.gitlab.com/orbit/remote/access/mcp/ ## **Enable it for your own namespace** Follow the steps bellow if you want to use it for another namespace than `gitlab-org`. 1. Navigate to your namespace. 2. Click on `Settings`, then on `Orbit`. You should get redirected to the Orbit setting page. 3. Then click on `Get Started` and then `Enable Indexing`. ## What to Try Start with something you're actually working on. The prompts below are ready to copy and paste - swap in your real file, function, or area. ### Get oriented fast > "I'm ramping up on \[system or area you're new to\]. Give me a high level map - key files, dependencies, and the call chain from the entry point down." > "What are the most most depended on modules in the monolith? Show me which files reference them most." ### Before you make a change > "I'm about to modify \[function or file\]. Show me every caller across all repos and give me a blast radius assessment." > "I'm deprecating \[endpoint or method\]. Find all internal callers - services, workers, anything that would break." > "Which tests are most likely to fail if I change \[file\]? Trace the call graph and surface the highest risk callers." ### Find before you build > "Before I write a new \[helper / service / pattern\], does GitLab already have something that does this? Show me all similar utilities and where they live." > "I need to \[describe what you're building\]. What existing patterns in the codebase should I follow or extend instead of writing from scratch?" ### Pipeline and CI/CD > "Which files in the monolith have the highest correlation with pipeline failures? Give me the top 10 and which jobs they tend to break." > "Show me the most frequently failing CI jobs in the last 30 days. For each one, what files changed in the MRs that introduced those failures?" > "Which CI jobs take the longest to run and what files most commonly trigger them? I want quick wins for pipeline cost reduction." ### For EMs and leads > "Give me a codebase health snapshot: which components have the most churn, the highest pipeline failure rate, and the fewest contributors?" > "Which parts of the codebase have the most coupling to other systems and the fewest owners? I want to find single points of failure on my team." --- ## Tell Us What You Find This is the most important part. One session with real signal is worth more than ten sessions we never hear about. **General Feedback, comment it HERE on this epic or add a child issue** :pray: - What did you ask? - Did you get a useful answer? - What surprised you - good or bad? - What didn't work? - UI/UX, usability opportunities _please test both WITH and WITHOUT Orbit, see how Duo Agent answers...and then how Orbit agent answers, so we get stronger signals of the difference this makes. Leave these comparisons as feedback in the epic._ **:rotating_light: CRITICAL BUGS --- add to the bug epic -\>** https://gitlab.com/groups/gitlab-org/-/work_items/21852 /cc @michaelangeloio @singhal.nitin