# Orbit Design Partner Feedback Welcome, design partners. This is your hub for feedback on Orbit (GitLab Knowledge Graph) during the closed beta. > [!note] > This epic and its child issues are **public**. Don't post anything you can't share externally: sensitive code, internal architecture, or customer data. For confidential reports, contact your GitLab DRI directly or use the private channels set up at your kickoff. ## What is Orbit? Orbit indexes your SDLC data (code, pipelines, MRs, issues, vulnerabilities) as a knowledge graph and exposes it to AI agents. Agents traverse relationships, not just files. A few examples of what becomes possible: - "What would break if I change this service?" - "Which tests correlate most with this file?" - "Find every caller of this function across all our repos." Available today on .com through: - **Duo Agent Platform** - the Planner Agent, Security Analyst Agent, Data Analyst Agent, CI Expert Agent, and Duo Developer Flow call Orbit when graph context helps answer the question. - **External agents** (Claude Code, Cursor, Codex, Duo CLI, etc.) via MCP at `https://gitlab.com/api/v4/orbit/mcp`. - **REST API**: `/api/v4/orbit/query`, `/api/v4/orbit/schema`, `/api/v4/orbit/status`, `/api/v4/orbit/tools`. Full docs: https://docs.gitlab.com/api/orbit/ ## What to try The most useful feedback comes from running the same real task through three different paths and telling us how the answers differ. Pick something you actually care about (a refactor you're planning, a service you're trying to understand, a pipeline you want to optimize) and run it through each setup below. ### 1. DAP without Orbit (your baseline) In the GitLab Duo Chat side panel, use the agent picker to select the Planner Agent, Security Analyst Agent, Data Analyst Agent, CI Expert Agent, or Duo Developer Flow. Ask them to help with your task while Orbit is disabled for your namespace. This is the agents-only baseline. Note what they can and cannot answer. ### 2. DAP with Orbit Enable Orbit on your namespace: **Settings → Orbit → Get Started → Enable Indexing**. Wait for indexing to complete (check status with `glab orbit remote status`). Run the same task through the same agents. The question we want answered: did Orbit change the answer? Was it more specific, more correct, more useful? Or did the agent ignore the graph entirely? ### 3. External agents via MCP Point your own agent (Claude Code, Cursor, Codex, Duo CLI, anything that speaks MCP) at Orbit: ```bash brew install glab # 1.94 or later glab auth login glab orbit remote status # should report 6 components healthy ``` 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`. Run the same task. How did your external agent compare to the DAP path? Where did it shine, where did it stumble? <details> <summary><strong>Prompts to get started</strong></summary> **Get oriented** - "Give me a high level map of [system]: key files, dependencies, call chains." - "What are the most depended on modules in our codebase?" **Before you change something** - "Show me every caller of [function] across all repos. Blast radius assessment." - "I'm deprecating [endpoint]. Find all internal callers." **Find before you build** - "Before I write a new [helper or service], do we have something similar?" **Pipeline and CI** - "Which files correlate most with pipeline failures? Top 10 with the jobs they tend to break." - "Which CI jobs take the longest? What files trigger them most?" **For EMs and leads** - "Codebase health snapshot: most churn, highest failure rate, fewest contributors." - "Where is coupling high and ownership thin? Find single points of failure." </details> ## What we want to hear Anything is fair game. The feedback that helps us most: - **Did it work?** Did you get the answer you needed? How did the quality compare to your alternatives? - **Did it scale?** Indexing time, query latency, freshness, completeness on your repos. - **Did it fit?** Where did it integrate cleanly into your workflow? Where did it not? - **What's missing?** Languages we don't index, entities we don't model, queries we can't answer. - **Bugs.** Anything that didn't behave as expected. - **"Huh, that's surprising" moments**, good or bad. The unscripted observations are the most valuable. ## How to give feedback Two ways, pick whichever is faster: 1. **Comment on this epic** for quick observations, questions, kudos, or anything that doesn't need its own thread. 2. **Open a child issue** in `gitlab-org/orbit/knowledge-graph` and link it back to this epic for bugs, feature requests, or anything that needs follow-up. If you're not sure where it fits, comment here. We will route it. ## What you can expect from us - We read everything within one business day. - We respond on every thread, even if the response is "thanks, logged" or "this is intentional, here is why." - Significant bugs get linked to internal tracking and prioritized. - We share back regularly across the program: what we heard, what we are shipping, what we are deferring. ## Program contacts - **Meg Corren** (`@mcorren`) - GTM PM, design partner program lead - **Angelo** (`@michaelangeloio`) - Tech Lead For your dedicated SA or CSM, see your kickoff email. --- Orbit is in beta. Expect rough edges. Tell us about them.