This article helps when building or querying the Vibgrate code map does not behave as expected. It is for developers using Vibgrate Graph commands such as vg show, vg ask, and vg impact.
Build the code map first
The graph queries depend on a built code map. Build or update it incrementally:
vg build
If a query like vg ask or vg impact returns nothing useful, the most common cause is that the map has not been built yet, or is stale.
Check freshness
Check how fresh the map is and how it compares to your working tree:
vg status
This reports node and edge counts and staleness. If it shows the map is stale, rebuild with vg build.
Slow or stalling queries
The first vg ask downloads a local model once into a shared cache. Precompute the semantic index so queries are instant and fully offline afterward:
vg embed
After the model is cached, vg ask and vg serve work without the network.
Keeping the map current for a team
If teammates see stale results, make the map committable and auto-updating so it stays in sync with commits:
vg share
This installs a pre-commit hook, a deterministic merge driver, and the right .gitignore entries.
Serving the map to an AI assistant
If the goal is to feed the map to an AI assistant, run Vibgrate AI Context locally:
vg serve
Use vg install to wire it into supported assistants such as Claude, Cursor, Windsurf, VS Code, Codex, and Gemini.
Related
- The
vg build,vg status, andvg askcapabilities (Vibgrate Graph) - Proxy and offline configuration
- Vibgrate AI Context via
vg serve