Ask Your Codebase Questions with vg ask

vg ask lets you query the code map in plain English. It uses hybrid lexical + structural + semantic search and returns a budget-bounded context block you can paste straight into any AI assistant. After the first run it works fully offline.

Prerequisites

• A built code graph from vg build

Steps

1. Build the graph

vg build

2. Ask a question

vg "how does authentication work in this codebase"

You can also use the explicit subcommand:

vg ask

Vibgrate searches the graph and assembles the most relevant nodes.

3. Understand the context block

vg ask returns a budget-bounded context block — it fits a token budget so it stays paste-ready. The block combines lexical matches, structural relationships, and semantic similarity, so it surfaces relevant code even when your wording does not match identifiers exactly.

4. Precompute embeddings for speed

To make the next vg ask instant, precompute the semantic index:

vg embed

The local ONNX model downloads once into a shared cache and then runs fully offline.

5. Use the context with an AI assistant

Paste the returned context block into your AI assistant as grounding, then ask your real question. Because the context is scoped to relevant code, the assistant answers with far less hallucination.

Verification

Confirm vg ask returns a context block referencing real nodes from your repository. If results are thin, rebuild with vg build and run vg embed to enable semantic ranking.

Next Steps

• Precompute the semantic index with vg embed
• Serve the graph to your AI assistant with vg serve
• Explain a specific result with vg show