| name | graph-onboard |
| description | Set up GraQle on a repository — scan it into a knowledge graph, verify graph health, and run the first graph-powered question. Use when GraQle is installed but no graqle.json exists yet, or when the user asks to onboard, scan, or index the repo with GraQle. |
Graph Onboard (GraQle)
Take a repository from zero to a queryable knowledge graph.
- Check the install — run
graq doctor in the shell. If graq is not found,
install with pip install graqle first.
- Scan —
graq scan repo . from the repository root. This builds graqle.json
(nodes for modules, functions, classes, endpoints and their dependency edges).
The first full scan is never blocked by licensing caps, whatever the repo
size (scan time itself scales with the repo).
- Configure the backend —
graqle.yaml controls which LLM provider powers
reasoning (Anthropic by default; bring your own key via environment variables).
- Verify —
graq_inspect for node/edge counts and hub nodes, and
graq_graph_health for embedding/cache status. If chunks are unembedded,
follow its recommendation before relying on semantic recall.
- First question —
graq_reason(question="What are the highest-risk modules in this repo and why?") to demonstrate graph-powered reasoning.
- Keep it growing — call
graq_learn after completing tasks, and let
graq init install git hooks so the graph stays in sync with the code.